Paper deep dive
UniAR: A Unified Framework for Autism Recognition Enhanced by Multi-View Prompt Learning
Lei Xin, Zeheng Wang, Jiayin Zhu, Shihong Huang, Fanhu Zeng, Changjiang Jiang, Dengbo He, Yutao Yue, Zhenglun Kong
Intelligence
Status: not_run | Model: - | Prompt: - | Confidence: 0%
Entities (0)
Relation Signals (0)
No relation signals yet.
Cypher Suggestions (0)
No Cypher suggestions yet.
Abstract
Abstract:Autism Spectrum Disorder (ASD) is a complex neurodevelopmental disorder for which early and accurate diagnosis is critical to improving long-term developmental outcomes. However, existing ASD recognition methods are often constrained by the scarcity of diagnostic text data, forcing them to rely mainly on visual analysis and limiting their ability to model clinically meaningful semantic reasoning. To address this challenge, we propose UniAR, a unified framework enhanced by multi-granularity prompt learning for robust ASD recognition under heterogeneous data variations. Specifically, UniAR leverages a large multimodal model to generate hierarchical diagnostic descriptions at the word, phrase, and sentence levels, compensating for the lack of paired clinical reports. To align the generated semantics with visual evidence, we further design a Mixture-of-Experts-based Multi-Scale Alignment Module, which dynamically matches vector-quantized visual prototypes with semantic representations at corresponding granularities. Extensive experiments on four benchmarks covering brain MRI and facial expression scenarios show that UniAR consistently outperforms existing state-of-the-art methods, achieving average accuracies of 75.9\% on MRI benchmarks and 91.6\% on facial benchmarks, while improving average Accuracy on MRI benchmarks by 1.5 percentage points and average Accuracy on facial benchmarks by 1.2 percentage points over baselines. These results demonstrate that UniAR offers a robust and interpretable framework for ASD screening under semantic scarcity.
Tags
Links
- Source: https://arxiv.org/abs/2609.31298v1
- Canonical: https://arxiv.org/abs/2609.31298v1
Trouble viewing inline? Open PDF directly →
Full Text
57,263 characters extracted from source content.
Expand or collapse full text
UniAR: A Unified Framework for Autism Recognition Enhanced by Multi-View Prompt Learning Conference: Proceedings of the 34th ACM International Conference on Multimedia; November 10–14, 2026; Rio de Janeiro, BrazilProceedings of the 34th ACM International Conference on Multimedia (M ’26), November 10–14, 2026, Rio de Janeiro, BrazilISBN: 979-8-4007-2213-4/2026/11DOI: 10.1145/3767308.3835308CCS: Computing methodologies Natural language generationCCS: Applied computing Health informatics Lei Xin Note: The first three authors contributed equally to this work. Affiliation: Wuhan University, Wuhan, Hubei, China email: 2835838600@q.com , Zeheng Wang Affiliation: Northeast Normal University, Changchun, Jilin, China email: wangzeheng624@nenu.edu.cn , Jiayin Zhu Affiliation: The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, Guangdong, China email: jzhu709@connect.hkustgz.edu.cn , Shihong Huang Affiliation: Nanjing Agricultural University, Nanjing, Jiangsu, China email: 23123122@stu.njau.edu.cn , Fanhu Zeng Affiliation: Imperial College London, London, United Kingdom email: challengezengfh@gmail.com , Changjiang Jiang Affiliation: Wuhan University, Wuhan, Hubei, China email: jiangcj@whu.edu.cn , Dengbo He Affiliation: The Hong Kong University of Science and Technology, Hong Kong, China email: dengbohe@ust.hk , Yutao Yue Affiliation: The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, Guangdong, China email: yutaoyue@hkust-gz.edu.cn and Zhenglun Kong Note: Zhenglun Kong is the corresponding author. Affiliation: Harvard University, Boston, USA email: zhenglun_kong@hms.harvard.edu © c Abstract. Autism Spectrum Disorder (ASD) is a complex neurodevelopmental disorder for which early and accurate diagnosis is critical to improving long-term developmental outcomes. However, existing ASD recognition methods are often constrained by the scarcity of diagnostic text data, forcing them to rely mainly on visual analysis and limiting their ability to model clinically meaningful semantic reasoning. To address this challenge, we propose UniAR, a unified framework enhanced by multi-granularity prompt learning for robust ASD recognition under heterogeneous data variations. Specifically, UniAR leverages a large multimodal model to generate hierarchical diagnostic descriptions at the word, phrase, and sentence levels, compensating for the lack of paired clinical reports. To align the generated semantics with visual evidence, we further design a Mixture-of-Experts-based Multi-Scale Alignment Module, which dynamically matches vector-quantized visual prototypes with semantic representations at corresponding granularities. Extensive experiments on four benchmarks covering brain MRI and facial expression scenarios show that UniAR consistently outperforms existing state-of-the-art methods, achieving average accuracies of 75.9% on MRI benchmarks and 91.6% on facial benchmarks, while improving average Accuracy on MRI benchmarks by 1.5 percentage points and average Accuracy on facial benchmarks by 1.2 percentage points over baselines. These results demonstrate that UniAR offers a robust and interpretable framework for ASD screening under semantic scarcity. Keywords: ASD Recognition, Affective Computing, Multimodal Learning, Multi-Granularity Prompt Learning, Cross-Modal Alignment †c-license: by 1. Introduction Autism Spectrum Disorder (ASD) represents a complex neuro-developmental condition characterized by core deficits in social communication and interaction, alongside restricted, repetitive patterns of behavior. According to statistics from the World Health Organization (WHO), the global average prevalence of ASD in children is approximately 1% (Salari et al., 2022). A substantial body of research indicates that early childhood, typically defined as 0–6 years, serves as a critical window for neurodevelopment; achieving early identification and intervention during this phase is instrumental in improving long-term developmental outcomes (Sandbank et al., 2023; Dawson et al., 2012; Santomauro et al., 2025). Consequently, enhancing the accessibility and accuracy of early ASD diagnosis holds significant social and public health implications (Santomauro et al., 2025). In clinical practice, screening for autism spectrum disorder (ASD) relies on standardized scales (e.g., M-CHAT-R/F) and parent reports (Robins et al., 2014). However, these subjective measurement methods often struggle to address the inherent heterogeneity of ASD presentation (Wieckowski et al., 2023; Robins et al., 2014; Hobson and Petty, 2021; Lord, 2019). Although machine learning has been applied to automatically identify ASD through behavioral imaging, existing methods primarily focus on unimodal visual analysis. Therefore, they overlook crucial semantic descriptors in clinical diagnosis, such as emotional responses and social interactions (Uddin et al., 2024). Figure 1. The framework comparisons.Existing autism recognition methods (a-b) are mainly implemented by domain experts or neural networks. In contrast, UniAR (c) integrates expert knowledge with neural networks to provide a more interpretable recognition solution, which is capable of outputting both diagnostic reports and results simultaneously.Comparison of clinical diagnosis, conventional black-box AI diagnosis, and the proposed multi-view prompting framework, which produces both an ASD prediction and interpretable evidence. While multimodal learning (Khattak et al., 2023; Zeng et al., 2025b) offers a promising avenue for integrating visual and semantic data, its application faces several obstacles. First, high-quality paired visual-text data is scarce and expensive to obtain (Wu et al., 2024). Furthermore, strict privacy regulations surrounding sensitive diagnostic text severely limit data sharing (Michelson et al., 2022; Ganggayah et al., 2025). These limitations prevent current multimodal methods from fully realizing their potential, thus hindering the deployment of accurate and practical diagnostic systems. To address the challenges of data limitations and missing text modalities, we propose UniAR, a Multi-Granularity Prompt Learning framework based on Cross-modal Alignment, as shown in Figure 1. Specifically, we utilize generative semantic augmentation to automatically synthesize multi-granularity descriptive text from visual inputs, effectively compensating for the scarcity of real diagnostic reports. Recognizing the complexity of ASD symptoms, we design a hierarchical “word-phrase-sentence” alignment mechanism that maps semantic descriptions to pixel-level, regional, and global visual features, respectively. By linking diagnostic semantics with behavioral biomarkers at different granularities, UniAR enables effective multi-scale visual-semantic fusion. We conduct extensive experimental validation of UniAR on four heterogeneous public benchmarks, covering brain MRI connectivity (ABIDE I & I) and facial behavioral expressions (HRM & Kaggle). While these public benchmarks mainly provide standard binary ASD-vs-control evaluation settings, they are limited in reflecting fine-grained symptom severity and real-world distribution shifts encountered in practical screening. To address this limitation, we further construct the first cross-platform social-media autism benchmark, termed ASD-M, which consists of autism-related videos paired with expert-associated diagnostic descriptions and formulates autism-related assessment as a four-class problem (i.e., Normal, Mild, Moderate, and Severe). Experimental results show that UniAR consistently outperforms strong single-modal and multimodal baselines, achieving average accuracies of 75.9% on MRI benchmarks and 91.6% on facial benchmarks. UniAR also achieves the best performance on the challenging ABIDE I and dynamic HRM datasets. Moreover, on the self-constructed ASD-M benchmark, UniAR consistently improves over the EAC baseline across all evaluated social platforms, further demonstrating its robustness and practical generalization. In summary, our paper makes the following contributions: • We propose UniAR, a unified visual-semantic alignment framework for robust ASD recognition that transforms low-level visual biomarkers into interpretable high-level diagnostic insights through multi-granularity prompting and semantic-guided modeling. • We develop a hierarchical word-phrase-sentence visual-semantic alignment mechanism that associates visual biomarkers with diagnostic semantics at multiple granularities, and further enhance visual representation learning through codebook-based prototype refinement, thereby improving robustness and diagnostic reasoning under limited data. • Extensive experiments on public MRI and facial expression benchmarks, as well as a cross-platform social-media evaluation setting, demonstrate that UniAR consistently outperforms strong baselines while showing promising robustness and semantic interpretability. 2. Related Work Figure 2. Overview of UniAR. The visual stream extracts multi-scale visual prototypes via a learnable visual codebook-augmented multi-hierarchical encoder. The semantic stream uses hierarchical prompts with GPT-4o to generate word-, phrase- and sentence-level descriptions, encoded by a frozen text encoder. A Scale-Aware Alignment Module aligns and fuses these multi-modal features for final classification. Pipeline of UniAR, including hierarchical visual encoding, AI-generated word-, phrase-, and sentence-level diagnostic descriptions, multi-scale cross-modal alignment, and final ASD classification. ASD Recognition. Automated ASD recognition can be broadly grouped into neuroimaging-based diagnosis, facial-behavior analysis, and emerging multimodal systems. In neuroimaging, recent review papers indicate that MRI-based ASD diagnosis remains a major research paradigm, with most studies focusing on stronger discriminative modeling of resting-state or structural biomarkers on datasets such as ABIDE (Uddin et al., 2024; Schielen et al., 2024; Ganggayah et al., 2025). Representative methods include connectome-based predictive modeling, graph-based modeling of functional connectivity, and more recent transformer or spectral architectures for ASD-related brain analysis (Shen et al., 2017; Kim et al., 2021; Li et al., 2021; Bedel et al., 2023; Duan et al., 2024; Duan et al., 2025; Liu et al., 2024). In parallel, facial and behavioral approaches identify ASD-related cues from head pose, facial expressions, and social-behavioral responses in images or videos (Liu et al., 2024; Deng et al., 2024; Ismail and Ali, 2025). Their implementations often build on robust FER techniques such as region attention, uncertainty modeling, and noisy-label learning (Wang et al., 2020b; Wang et al., 2020a; She et al., 2021; Zhang et al., 2021; Zhang et al., 2022). Some work further explores multimodal evidence such as audio-visual interaction modeling (Zhao et al., 2025b). Prior ASD-specific methods can be summarized as connectome modeling, behavioral cue modeling, and multimodal fusion, yet they mainly strengthen discriminative representations while explicit clinical-semantic modeling remains limited. Semantic Reconstruction and Missing-Modality Learning. Another related line addresses limited supervision through augmentation, modality completion, or prompt-based adaptation. Classical augmentation improves generalization by enriching the sample space (Shorten and Khoshgoftaar, 2019), and recent neuroimaging work has extended this idea to fMRI-specific augmentation and computation toolchains (Fang et al., 2025). In ASD settings, ASD-GANNet and GARL enhance robustness by synthesizing pseudo-samples or enriched feature distributions for small-data regimes (Khan and Shang, 2024; Zhou et al., 2024). More broadly, missing-modality learning studies how to preserve performance when one or more modalities are absent, typically through imputation, distillation, or robust fusion (Wu et al., 2024; Wang et al., 2023). Prompt-based multimodal adaptation (Khattak et al., 2023; Zeng et al., 2025a) further shows that learnable prompts can improve cross-modal conditioning and transfer (Guo et al., 2025; Zhao et al., 2025a). DeFakerOne (Team, 2026) introduces a industrial-grade image detection framework, demonstrating the feasibility leveraging massive datasets. Taken together, these methods can be summarized as sample-space augmentation, feature or modality completion, and prompt-conditioned adaptation. However, they mainly target robustness under incomplete inputs or limited labels, whereas diagnosis-level semantic reconstruction under missing clinical text remains underexplored. Fake-HR1 (Jiang et al., 2026b) and DS-Net (Qu et al., 2026) is based on question difficulty and thequality of dataset to achieve adaptive perception training. Medical Visual-Language Alignment. Medical vision-language learning has progressed from coarse report-level matching to increasingly fine-grained semantic alignment. At a conceptual level, multi-grained vision-language pretraining showed the value of matching textual units with visual concepts across granularities (Zeng et al., 2021). In the medical domain, GLoRIA aligns report words with image subregions for label-efficient representation learning; MGCA models region-, instance-, and disease-level correspondences between medical images and reports; and MedKLIP introduces structured medical knowledge to align entity-level descriptions with image patches for diagnosis and grounding (Huang et al., 2021; Wang et al., 2022; Wu et al., 2023). More recent work extends this trend to brain-domain vision-language pretraining and aspect-level disease semantic decomposition (Monajatipoor et al., 2024; Phan et al., 2024). Ivy-Fake (Jiang et al., 2026a) and TwSG (Jiang et al., 2026c) pioneered an explainable dataset for image understanding. Overall, existing methods can be summarized as report-level matching, multi-granularity local-global alignment, and knowledge-guided semantic grounding. However, they generally assume that paired reports or curated disease descriptions are available during training, leaving report-absent medical settings insufficiently studied. 3. The Proposed Method 3.1. Preliminaries ASD recognition often suffers from semantic scarcity: visual observations are available, while fine-grained clinical descriptions are usually missing. UniAR addresses this problem by modeling visual evidence and the latent diagnostic semantics inferred from it. Problem Formulation. We formulate ASD recognition as a supervised classification problem. Let =(xi,yi)i=1ND=\(x_i,y_i)\_i=1^N denote a dataset, where xi∈x_i is the visual input and yi∈y_i is the diagnostic label. Since explicit clinical semantics are typically unavailable, we approximate them using a large multimodal generator G and optimize θ∗=argmax∑i=1NθlogP(yi∣xi,(xi);θ)θ^*= _θ _i=1^N P (y_i x_i,G(x_i);θ ), where θ denotes the learnable parameters of UniAR. Loss Function. The overall objective combines classification, semantic alignment, hierarchical regularization, fusion quality, and prototype learning: ℒtotal=ℒcls+αℒalign+βℒsep+γℒfusion+λvℒvqL_total=L_cls+ _align+ _sep+ _fusion+ _vL_vq, where ℒclsL_cls supervises ASD prediction, ℒalignL_align enforces cross-modal consistency, ℒsepL_sep preserves complementary hierarchical cues, ℒfusionL_fusion regularizes multi-scale fusion, and ℒvqL_vq stabilizes prototype learning. Based on this formulation, UniAR is designed to reconstruct missing diagnostic semantics from visual evidence and align them with multi-scale visual representations for robust ASD recognition. 3.2. Framework Overview ASD recognition under semantic scarcity involves three coupled challenges. First, raw visual observations are highly sensitive to subject-specific appearance variations, acquisition noise, and site-specific heterogeneity, making the learned representations unstable across samples. Second, although visual data may contain discriminative behavioral biomarkers, explicit clinical semantic supervision is typically unavailable, preventing the model from linking visual evidence to interpretable diagnostic concepts. Third, even when semantic descriptions are constructed, visual and textual representations remain mismatched in both feature distribution and abstraction level, making direct multimodal fusion suboptimal. To address these issues, we propose UniAR, a unified visual-semantic framework for ASD recognition, as illustrated in Figure 2. The framework follows a three-stage pipeline. First, a visual representation construction stage transforms raw visual inputs into prototype-based multi-scale features through a hierarchical encoder and a learnable visual codebook, thereby reducing instance-specific noise while preserving ASD-relevant patterns. Second, a semantic construction stage leverages a large multimodal model to generate hierarchical diagnostic descriptions at the word, phrase, and sentence levels, providing explicit semantic anchors for otherwise purely visual evidence. Third, a scale-aware alignment stage progressively aligns visual prototypes with semantic embeddings at matched granularities, enabling robust cross-modal interaction and final ASD prediction. In this way, UniAR does not treat multimodal learning as a simple combination of visual and textual features. Instead, it first stabilizes the visual representation, then reconstructs missing diagnostic semantics, and finally resolves the scale mismatch between the two modalities through explicit alignment. This design allows the framework to jointly improve classification robustness and diagnostic interpretability. 3.3. Multi-Granularity Visual-Semantic Representation Construction Module Following the framework overview, we first construct the multi-granularity visual-semantic representation used by UniAR. Effective ASD recognition requires both robust visual representations and semantically grounded diagnostic cues. However, raw visual features are often entangled with subject-specific variations and irrelevant noise, while stabilized visual patterns still lack explicit clinical meaning. We therefore construct the visual-semantic representation in three stages: multi-granularity representation space, codebook-enhanced Hierarchical Visual Encoding and prompt-driven multi-granularity semantic generation. Multi-Granularity Representation Space. To reflect the hierarchical nature of ASD evidence, both modalities are represented at multiple granularities. For the visual modality, the encoder ℰvisE_vis produces: (1) =ℰvis(x)=(l)∈ℝHl×Wl×D∣l=1,…,L,V=E_vis(x)=\v^(l) ^H_l× W_l× D l=1,…,L\, where (l)v^(l) denotes the visual representation at scale l. For the semantic modality, the generator G with hierarchical prompts P yields: (2) =(x,)=Tk∣k∈word,phrase,sent.T=G(x,P)=\T_k k∈\word,phrase,sent\\. Here, TkT_k denotes the generated diagnostic description at semantic granularity k. This formulation allows UniAR to associate visual evidence with semantic concepts at matched granularities. Codebook-Enhanced Hierarchical Visual Encoding. Directly aligning raw visual features with language is unreliable due to identity-specific appearance, background interference, and acquisition variation. We therefore first transform raw observations into more stable visual prototypes. Given a visual observation, a multi-hierarchical encoder extracts feature maps at multiple scales: =vll=1L,vl∈ℝHl×Wl×DV=\v_l\_l=1^L, v_l ^H_l× W_l× D, where vlv_l denotes the feature representation at scale l. To suppress redundant variations, we introduce a learnable Visual Codebook =kk=1KC=\c_k\_k=1^K, where each codeword k∈ℝDc_k ^D represents an ASD-relevant visual prototype. For each spatial feature vector vl(i,j)v_l^(i,j), nearest-neighbor quantization is performed as: (3) zq(vl(i,j))=k,wherek=argminn∈1,…,K‖vl(i,j)−n‖2.z_q(v_l^(i,j))=c_k, k= _n∈\1,…,K\\|v_l^(i,j)-c_n\|_2. zq(⋅)z_q(·) denotes the quantization operator. By mapping continuous features into a discrete prototype space, the model suppresses instance-specific noise and encourages visually diverse samples to share ASD-salient behavioral patterns. Gradients are propagated through the quantization operation using the straight-through estimator. Prompt-Driven Multi-Granularity Semantic Generation. Although the visual codebook stabilizes the visual stream, the resulting prototypes remain diagnostically under-specified. To provide interpretable diagnostic anchors, we further construct hierarchical semantics from the visual input. Because real ASD datasets rarely contain dense reports, we employ GPT-4o to generate descriptions at three complementary granularities: =Tword,Tphrase,TsentT=\T_word,T_phrase,T_sent\, where TwordT_word, TphraseT_phrase, and TsentT_sent denote word-level, phrase-level, and sentence-level descriptions, respectively. These levels progressively capture isolated attributes, local behavioral interactions, and holistic diagnostic context. The generated texts are encoded by a frozen text encoder: (4) g=Enc(Tg),g∈word,phrase,sent,S_g=Enc(T_g), g∈\word,phrase,sent\, yielding semantic embeddings wordS_word, phraseS_phrase, and sentS_sent. This hierarchical construction provides richer semantic anchors than a single global description and supports subsequent fine-grained cross-modal alignment. Figure 3. Detailed illustration of the Multi-Scale Alignment Module (MSAM) designed to tackle feature inconsistencies.Three-stage multi-scale alignment module showing semantic and visual features aligned through cross-modal attention and fused for prediction. 3.4. Multi-Scale Alignment Module After prototype refinement and semantic generation, the model obtains more stable visual patterns and more informative diagnostic descriptions. However, these two modalities are still not directly compatible. In particular, visual prototypes and semantic embeddings differ not only in modality, but also in abstraction level: local visual cues should not interact with sentence-level semantics in the same manner as global behavioral patterns. As a result, naive concatenation or direct fusion may introduce severe cross-modal mismatch and weaken the benefit of hierarchical semantics. To address this issue, we propose a Multi-Scale Alignment Module (MSAM) that explicitly performs scale-aware feature modulation and cross-modal interaction. Cross-Modal Alignment Objective. Since visual and semantic features come from heterogeneous spaces, UniAR learns a joint representation by minimizing their latent discrepancy: (5) min∑l,kθDϕ(fv((l)),ft((k))), _θ _l,kD_φ (f_v(v^(l)),f_t(t^(k)) ), where fv(⋅)f_v(·) and ft(⋅)f_t(·) are modality-specific projection functions, and DϕD_φ measures their discrepancy. This motivates the scale-aware alignment design by encouraging matched visual and semantic representations to become more compatible in the latent space. Scale-Adaptive Modulation with MoE. Before multimodal fusion, the model should first reduce the feature discrepancy between different granularities. To this end, we introduce a scale-adaptive modulator implemented with a Mixture-of-Experts (MoE) mechanism. The main idea is to decompose each input feature into shared and scale-specific components, so that common diagnostic structure can be preserved while scale-dependent characteristics are selectively emphasized. Let inx_in denote the input feature for a specific alignment branch, which can be either a visual representation invx_in^v or a semantic representation insx_in^s. We employ a Shared Expert ℰs(⋅)E_s(·) to model common patterns across different scales, and a Scale-Specific Expert ℰd(k)(⋅)E_d^(k)(·) for the k-th granularity. The modulated feature is computed as: (6) mod=ℰs(in)+λ⋅ℰd(k)(in),x_mod=E_s(x_in)+λ·E_d^(k)(x_in), where λ is a learnable gating factor. Both experts are implemented as MLPs. This design enables the model to preserve cross-scale commonality while adaptively highlighting the semantic attributes that are most relevant to each granularity, thereby producing more compatible representations for subsequent alignment. Table 1. Comparative experiments on MRI scenarios. Method ABIDE ABIDE I AVG Acc Recall Pre AUC Acc Recall Pre AUC Acc Recall Pre AUC CPM 61.4 58.3 53.3 64.7 60.3 61.0 58.5 62.0 60.9 59.7 55.9 63.4 Braingnn 60.2 60.4 50.3 58.2 58.2 59.3 50.2 57.3 59.2 59.9 50.3 57.8 SpectBGNN 59.6 67.3 51.9 61.4 69.6 71.0 72.2 71.4 64.6 69.2 62.1 66.4 STAGIN 68.5 71.2 69.2 74.1 68.4 67.8 69.1 72.0 68.5 69.5 69.2 73.1 Bolt 67.2 68.4 66.3 71.3 70.2 70.5 71.4 75.0 68.7 69.5 68.9 73.2 Causality 69.7 54.3 73.4 74.2 71.7 65.7 74.2 75.3 70.7 60.0 73.8 74.8 BrainNetMLP 76.8 77.0 76.5 80.2 71.5 68.5 72.5 74.8 74.2 72.8 74.5 77.5 GPT-4o 77.5 78.2 77.6 81.3 71.7 68.7 72.7 74.6 74.4 73.5 75.2 78.0 Ours 78.5 79.2 78.8 82.5 73.3 69.4 73.7 76.7 75.9 74.3 76.3 79.6 Table 2. Comparative experiments on facial expression scenarios. Method HRM Kaggle AVG Acc Recall Pre AUC Acc Recall Pre AUC Acc Recall Pre AUC RAN 78.4 73.5 75.8 81.0 98.2 97.8 97.5 99.1 88.3 85.7 86.7 90.1 SCN 79.3 74.8 76.5 82.5 94.5 93.2 94.8 95.5 86.9 84.0 85.7 89.0 DMUE 80.5 77.1 77.5 83.9 97.1 96.5 97.8 98.2 88.8 86.8 87.7 91.1 RUL 80.8 76.9 78.4 84.2 97.6 96.8 98.1 98.5 89.2 86.9 88.3 91.4 EAC 81.2 77.5 78.9 84.8 96.3 95.9 96.7 97.5 88.8 86.7 87.8 91.2 GPT-4o 82.6 79.2 80.3 85.7 98.2 97.2 97.8 98.7 90.4 88.2 89.1 92.1 Ours 84.5 81.2 81.8 87.6 98.7 97.9 98.2 98.7 91.6 89.6 90.0 93.2 Multi-Scale Cross-Modal Fusion. Once the features are modulated into a more compatible latent space, the next challenge is to establish effective interactions between matched visual and semantic granularities. We therefore adopt a multi-scale cross-attention mechanism to inject diagnostic semantics into visual prototypes in a scale-aware manner. For each branch, the modulated visual features are treated as Queries, while the corresponding semantic embeddings serve as Keys and Values. The cross-modal interaction is defined as: (7) Attention(Q,K,V)=Softmax(modv(mods)Tdk)mods,Attention(Q,K,V)=Softmax ( x_mod^v(x_mod^s)^T d_k )x_mod^s, where modvx_mod^v and modsx_mod^s denote the modulated visual and semantic features, respectively. Based on this interaction, the aligned representation is obtained by: (8) =LN(inv+ins+Attention(Q,K,V)),Z=LN (x_in^v+x_in^s+Attention(Q,K,V) ), where LN(⋅)LN(·) denotes Layer Normalization. This process progressively injects hierarchical diagnostic semantics into the visual stream, allowing the model to connect local visual biomarkers with fine-grained linguistic cues and global visual patterns with holistic diagnostic context. Finally, the aligned outputs from all branches are aggregated to form the final representation for ASD recognition. 4. Experiments 4.1. Experimental Setting Public Benchmarks. Autism-related biomarkers in our study are manifested in two complementary modalities, i.e., brain MRI connectivity and facial behavioral expressions. To comprehensively evaluate the robustness and generalization of UniAR, we conduct experiments on four public benchmarks together with one self-constructed cross-platform social-media benchmark. MRI Scenario (ABIDE I & I): We evaluate on the Autism Brain Imaging Data Exchange benchmarks. ABIDE I contains 1,112 subjects (539 ASD and 573 controls) from 17 sites, serving as a standard multi-site benchmark for ASD identification. ABIDE I includes 1,114 subjects from 19 sites and provides a more challenging setting with greater inter-site variability. Facial Expression Scenario (HRM & Kaggle): We use two facial benchmarks to cover both temporal and static behavioral cues. The HRM dataset contains 131,758 frames from 1,535 videos for evaluating subtle temporal facial dynamics. In addition, we use an aggregated static facial benchmark collected from Kaggle and Zenodo, containing 12,381 images, to assess generalization. ASD-M Dataset Construction and Evaluation Protocol.To complement the closed binary ASD-vs-control settings of existing public benchmarks, we further construct the first cross-platform social-media autism benchmark, termed ASD-M, for finer-grained autism assessment under realistic distribution shifts. ASD-M consists of autism-related videos paired with expert-associated diagnostic descriptions and defines autism-related assessment as a four-class problem with labels Normal, Mild, Moderate, and Severe. The videos are collected from three mainstream social-media platforms, namely Bilibili, YouTube, and TikTok. After data cleaning and screening, representative frames are manually extracted from each video for evaluation. All samples are annotated by three clinical practitioners, and each sample is associated with an expert-written diagnostic description covering three dimensions: social competence, language expression, and behavior. Category statistics of ASD-M are summarized in Figure 5. In experiments, ASD-M serves as an external validation benchmark for both cross-platform four-class classification and quantitative interpretability evaluation. Baselines. We compare UniAR against representative methods designed for the two visual scenarios.For MRI scenarios, we include methods from six paradigms: (i) graph-based learning, including BrainGNN (Li et al., 2021), STAGIN (Kim et al., 2021), and Spectral BGNN (Duan et al., 2024); (i) temporal modeling, represented by BolT (Bedel et al., 2023); (i) classical statistical prediction, represented by CPM (Shen et al., 2017);(iv) causal modeling, represented by the causality-inspired LSTM framework (Duan et al., 2025);and(v) MLP-based modeling, represented by BrainNetMLP (Hou et al., 2025),an efficient baseline for functional brain network classification. For facial expression scenarios, we compare with robust CNN-based, vision-language, and instruction-tuned methods, including SCN (Wang et al., 2020a), DMUE (She et al., 2021), RUL (Zhang et al., 2021), EAC (Zhang et al., 2022), RAN (Wang et al., 2020b). For the cross-platform ASD-M benchmark, we compare with EAC under the same eval protocol. Metrics. For the public classification benchmarks, we report Accuracy, Precision, Recall, and Area Under the ROC Curve (AUC), which measure diagnostic correctness and discriminative ability. For the ASD-M benchmark, we report Accuracy, F1-score, and Recall for four-class classification performance. To evaluate the quality of generated diagnostic descriptions, we further adopt ROUGE-1, ROUGE-2, and ROUGE-L against expert-written reports. Implementation Details. Our framework is implemented in PyTorch and trained on a single NVIDIA A100 GPU. For ABIDE I and I, we extract functional connectivity patterns as model input, while the facial datasets undergo standard visual preprocessing. Multi-Granularity prompts are generated using GPT-4o and then encoded by a frozen BERT encoder. For visual discretization, we use a multi-scale vector quantization module with a codebook size of 128 and an embedding dimension of 256. The quantized representations are projected back to the original feature space through a linear layer for dimensional consistency. In the MoE branch, each routing branch contains 4 experts, with hidden dimension 512 and input/output dimension 256. The gating network uses a 64-dimensional hidden layer to generate softmax routing coefficients. We train the model using the Adam optimizer with a learning rate of 1×10−41× 10^-4 and a batch size of 32 for 30 epochs, with early stopping. For the public benchmarks, the train/test ratio is set to 7:3. 4.2. Comparison Results Performance on MRI Scenarios. Table 1 reports the results on the heterogeneous ABIDE benchmarks. UniAR achieves the best overall performance on both ABIDE and ABIDE I, ranking first on all four metrics for ABIDE and obtaining the best Accuracy and AUC on ABIDE I. Compared with ASD-specific baselines, including CPM, Braingnn, SpectBGNN , STAGIN, Bolt and Causality, the advantage of UniAR is clear. In particular, relative to the causality-based model, UniAR improves Recall from 54.3% to 79.2% on ABIDE and from 60.0% to 74.3% on the average results, showing a much better balance between sensitivity and precision for screening-oriented ASD recognition. We further include BrainNetMLP and GPT-4o as stronger generic and foundation-model baselines, and UniAR still remains superior in overall Accuracy and AUC, indicating that the gain does not come merely from generated semantics, but from the joint effect of the multi-granularity visual-semantic representation construction module and the multi-scale alignment module. Performance on Facial Expression Scenarios. Table 2 reports the results on the facial expression benchmarks. UniAR consistently achieves the best overall performance across both dynamic and static settings. Compared with ASD/facial-specific baselines, including RAN, SCN, DMUE, RUL, and EAC, the advantage of UniAR is clear on both HRM and Kaggle. In particular, on the dynamic HRM benchmark, UniAR reaches 84.5% Accuracy, outperforming EAC by 3.3 percentage points and GPT-4o by 1.9 percentage points, while also ranking first on Recall, Precision, and AUC. On the average results, UniAR improves Accuracy from 88.8% to 91.6% over EAC and from 90.4% to 91.6% over GPT-4o. These results suggest that the gain of UniAR does not come solely from generated descriptions, but from the joint effect of the multi-granularity visual-semantic representation construction module and the multi-scale alignment module. The consistent gains across HRM and Kaggle further indicate that the proposed framework is effective for ASD-related facial behavior understanding across both dynamic and static scenarios. Table 3. Ablation study on MRI scenarios. Method Modules ABIDE ABIDE I AVG Word Phrase Sent. Fusion Acc Recall Pre AUC Acc Recall Pre AUC Acc Recall Pre AUC Baseline × × × × 76.2 77.1 76.5 80.1 69.1 65.8 70.2 72.3 72.7 71.5 73.4 76.2 w/ Word ✓ × × × 76.9 77.6 77.2 80.7 70.0 66.9 71.5 73.1 73.5 72.3 74.4 76.9 w/ Word, Phrase ✓ ✓ × × 77.6 78.3 78.0 81.4 71.0 68.1 72.6 74.2 74.3 73.2 75.3 77.8 w/ Word, Phrase, Sentence ✓ ✓ ✓ × 78.1 78.8 78.5 82.0 71.8 68.8 73.2 75.0 75.0 73.8 75.9 78.5 Ours ✓ ✓ ✓ ✓ 78.5 79.2 78.8 82.5 73.3 69.4 73.7 76.7 75.9 74.3 76.3 79.6 Table 4. Experimental results of different codebook sizes on multiple datasets. Codebook Size HRM Kaggle ABIDE ABIDE I Accuracy Recall Precision AUC Accuracy Recall Precision AUC Accuracy Recall Precision AUC Accuracy Recall Precision AUC 64 83.8 80.5 81.1 86.9 98.4 97.6 97.9 98.1 78.1 78.6 78.2 81.9 71.8 68.9 73.2 75.2 128 84.5 81.2 81.8 87.6 98.7 97.9 98.2 98.7 78.5 78.9 78.8 82.5 72.3 69.4 73.7 75.7 256 84.2 81.4 81.5 87.5 98.5 97.7 98.1 98.8 78.3 78.9 78.6 82.6 72.1 69.3 73.5 75.8 Table 5. Comparative experiments with different numbers of experts on multiple datasets. Number of Experts HRM Kaggle ABIDE ABIDE I Accuracy Recall Precision AUC Accuracy Recall Precision AUC Accuracy Recall Precision AUC Accuracy Recall Precision AUC 2 84.2 80.9 81.5 87.2 98.4 97.7 97.9 98.4 78.2 78.8 78.6 82.2 72.2 69.2 73.5 75.3 3 84.4 81.2 81.7 87.4 98.7 97.7 98.1 98.6 78.4 78.7 78.9 82.3 72.1 69.2 73.6 75.5 4 84.5 81.2 81.8 87.6 98.7 97.9 98.2 98.7 78.5 79.2 78.8 82.5 72.3 69.4 73.7 75.7 4.3. Ablation Study Figure 4. Qualitative visualization of attention mechanisms and hierarchical semantic alignment. We conduct ablation and design analysis from two perspectives. First, we examine the effectiveness of the proposed hierarchical alignment and fusion mechanism by progressively adding the word-level, phrase-level, sentence-level, and fusion modules. Second, we study whether two key architectural choices—the codebook size and the number of experts in the MoE branch—are reasonably selected. These aim to verify not only that the proposed components are effective, but also that the overall design of UniAR is well aligned with the demands of ASD recognition under semantic scarcity. Module Ablation on MRI Scenarios. We first evaluate the contribution of each semantic alignment component and the fusion module on the MRI benchmarks, as shown in Table 3. Starting from the baseline (w/o All), performance improves steadily as word-level, phrase-level, and sentence-level alignment are introduced in sequence. This trend suggests that hierarchical semantic guidance progressively enriches the model’s understanding of ASD-related patterns, moving from fine-grained local semantics to more holistic diagnostic context. In particular, the addition of sentence-level alignment brings a clear improvement over the word+phrase setting, indicating that global semantic context is helpful for interpreting heterogeneous MRI biomarkers. After further adding the fusion module, the full model achieves the best overall performance, which supports the role of fusion in integrating multi-scale visual-semantic evidence. Similar observations can also be found on the facial expression benchmarks. Sensitivity to Codebook Size. We next investigate the influence of codebook size on UniAR, as summarized in Table 4. Since the visual codebook is responsible for quantizing visual features into more stable prototypes, its size directly affects the balance between representation diversity and prototype compactness. The results show that a codebook size of 128 achieves the best overall performance across the four datasets. A smaller codebook (64) appears insufficient to capture diverse ASD-related visual patterns, while a larger one (256) introduces limited or inconsistent gains, suggesting that excessive prototype capacity may weaken representation compactness and stability. These results support the choice of a moderate codebook size for robust multi-scale visual abstraction. Sensitivity to the Number of Experts. We further analyze the effect of the number of experts in the MoE-based modulation branch in Table 5, which shows that increasing the number of experts from 2 to 4 consistently improves performance across most datasets, and the best overall results are achieved with 4 experts. This suggests that a moderate increase in expert diversity is beneficial for modeling heterogeneous visual-semantic patterns, especially when aligning features from different abstraction levels. In contrast, smaller expert configurations provide limited routing flexibility and are less effective at handling the complexity of cross-dataset variations. Figure 5. Data volume statistics of different categories across platforms.Pie charts showing the distribution of normal, mild, moderate, and severe samples across Bilibili, YouTube, and TikTok. 4.4. Qualitative Analysis and Visualization To qualitatively examine the interpretability of UniAR, we use Grad-CAM to project decision activations back to the input space, where warmer colors indicate higher relevance. As shown in Figure 4, UniAR exhibits two visible advantages over the EAC baseline. First, its high-response regions are more concentrated on clinically relevant facial areas, such as the periorbital and perioral regions, with substantially less attention leakage to irrelevant background regions. Second, the activation patterns of UniAR appear more spatially coherent, whereas the baseline shows more scattered and fragmented responses. These observations suggest that UniAR establishes a more reliable correspondence between facial evidence and diagnostically relevant semantics. These qualitative patterns are also consistent with the design of our hierarchical visual-semantic alignment. Word-level prompts provide fine-grained semantic anchors for local facial cues, phrase-level prompts capture relations among facial components and behavioral patterns, and sentence-level prompts introduce more global diagnostic context. Although Grad-CAM does not provide direct causal evidence for each semantic level, the improved localization and coherence of UniAR support the effectiveness of aligning visual evidence with multi-granularity diagnostic semantics. 4.5. Cross-Platform Quantitative Validation and Interpretable Analysis To further evaluate the practical generalization of UniAR beyond closed public benchmarks, we conduct additional experiments on the self-constructed ASD-M benchmark. In particular, this setting is designed to assess two aspects: whether UniAR remains robust under cross-platform distribution shifts, and whether the generated diagnostic descriptions remain semantically consistent with expert-associated annotations in realistic scenarios. Cross-Platform Four-Class Classification. We first evaluate UniAR on ASD-M under the four-class setting of Normal, Mild, Moderate, and Severe. As shown in Figure 6, UniAR consistently outperforms the EAC baseline across the evaluated social-media platforms in terms of Accuracy, F1-score, and Recall. These results indicate that the proposed visual-semantic alignment strategy generalizes more robustly than a strong task-specific baseline under realistic cross-platform distribution shifts. Quantitative Interpretability Evaluation. Since the diagnostic descriptions used in UniAR are generated rather than directly observed, it is important to examine whether they remain semantically faithful to expert knowledge instead of introducing unreliable hallucinated content. To this end, we further compare the generated descriptions of UniAR with expert-associated diagnostic reports using ROUGE-1, ROUGE-2, and ROUGE-L. As reported in Table 6, UniAR achieves favorable semantic consistency with expert reports across all evaluated platforms. These results suggest that the generated diagnostic descriptions capture diagnostically meaningful content to a reasonable extent, thereby providing additional quantitative support for the interpretability of UniAR in realistic deployment settings. Figure 6. Cross-platform performance comparison on social media datasets.Bar charts comparing UniAR and the EAC baseline on accuracy, F1-score, and recall across three social-media platforms. Table 6. Evaluation metrics on cross-platform datasets. Platform Rouge-1 Rouge-2 Rouge-L Bilibili 72.3 65.7 70.1 YouTube 78.6 72.4 76.9 TikTok 77.8 71.3 76.2 5. Conclusion This paper addresses the challenge of semantic scarcity in ASD recognition by proposing UniAR, a unified framework that reconstructs hierarchical diagnostic semantics through multi-granularity prompt learning and aligns them with visual evidence in a scale-aware manner. Extensive experiments show that structured semantic supervision is highly beneficial for robust ASD recognition, especially under heterogeneous MRI settings, subtle facial behavior variations, and incomplete diagnostic text conditions. By combining prototype-based visual representation refinement with hierarchical visual-semantic alignment, UniAR effectively improves both robustness and interpretability, providing a practical way to bridge low-level biomarkers and high-level diagnostic reasoning. References Bedel et al. (2023) H. A. Bedel, I. Sivgin, O. Dalmaz, S. U. Dar, and T. Çukur BolT: fused window transformers for fmri time series analysis. Medical image analysis 88, p. 102841. Cited by: §2, §4.1. Dawson et al. (2012) G. Dawson, E. J. Jones, K. Merkle, K. Venema, R. Lowy, S. Faja, D. Kamara, M. Murias, J. Greenson, J. Winter, et al. Early behavioral intervention is associated with normalized brain activity in young children with autism. Journal of the American Academy of Child & Adolescent Psychiatry 51 (11), p. 1150–1159. Cited by: §1. Deng et al. (2024) A. Deng, T. Yang, C. Chen, Q. Chen, L. Neely, and S. Oyama Language-assisted deep learning for autistic behaviors recognition. Smart Health 32, p. 100444. Cited by: §2. Duan et al. (2024) P. Duan, N. C. Dvornek, J. Wang, J. Eilbott, Y. Du, D. G. Sukhodolsky, and J. S. Duncan Spectral brain graph neural network for prediction of anxiety in children with autism spectrum disorder. In 2024 IEEE International Symposium on Biomedical Imaging (ISBI), p. 1–5. Cited by: §2, §4.1. Duan et al. (2025) P. Duan, N. C. Dvornek, J. Wang, L. H. Staib, and J. S. Duncan Causal modeling of fmri time-series for interpretable autism spectrum disorder classification. In 2025 IEEE 22nd International Symposium on Biomedical Imaging (ISBI), p. 1–5. Cited by: §2, §4.1. Fang et al. (2025) Y. Fang, J. Zhang, L. Wang, Q. Wang, and M. Liu ACTION: augmentation and computation toolbox for brain network analysis with functional mri. NeuroImage 305, p. 120967. Cited by: §2. Ganggayah et al. (2025) M. D. Ganggayah, D. Zhao, E. J. Y. Liew, N. A. M. Nor, T. Paramasivam, Y. Y. Lee, N. I. A. Hasan, and S. Shaharuddin Accelerating autism spectrum disorder care: a rapid review of data science applications in diagnosis and intervention. Asian Journal of Psychiatry 108, p. 104498. Cited by: §1, §2. Guo et al. (2025) H. Guo, F. Zeng, Z. Xiang, F. Zhu, D. Wang, X. Zhang, and C. Liu Hide-llava: hierarchical decoupling for continual instruction tuning of multimodal large language model. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), p. 13572–13586. Cited by: §2. Hobson and Petty (2021) H. Hobson and S. Petty Moving forwards not backwards: heterogeneity in autism spectrum disorders. Molecular Psychiatry 26 (12), p. 7100–7101. Cited by: §1. Hou et al. (2025) J. Hou, Z. Song, and E. E. Kuruoglu BrainNetMLP: an efficient and effective baseline for functional brain network classification. In International Workshop on Efficient Medical Artificial Intelligence, p. 11–20. Cited by: §4.1. Huang et al. (2021) S. Huang, L. Shen, M. P. Lungren, and S. Yeung Gloria: a multimodal global-local representation learning framework for label-efficient medical image recognition. In Proceedings of the IEEE/CVF international conference on computer vision, p. 3942–3951. Cited by: §2. Ismail and Ali (2025) W. N. Ismail and M. A. Ali Multiscale attention-based network to enhance detection and classification of autism spectrum disorders using convolutional neural network. PeerJ Computer Science 11, p. e3134. Cited by: §2. Jiang et al. (2026a) C. Jiang, W. Dong, Z. Zhang, F. Yu, W. Peng, X. Yuan, Y. Bi, M. Zhao, Z. Zhou, C. Si, and C. Shan Ivy-fake: a unified explainable framework and benchmark for image and video aigc detection. In Proceedings of the 2026 International Conference on Multimedia Retrieval, ICMR ’26, p. 2438–2447. External Links: ISBN 9798400726170, Link, Document Cited by: §2. Jiang et al. (2026b) C. Jiang, X. Sha, F. Yu, J. Liu, J. Liu, M. Fang, C. Zhang, and W. Lu Fake-hr1: rethinking reasoning of vision language model for synthetic image detection. In ICASSP 2026 - 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), p. 10482–10486. External Links: Document Cited by: §2. Jiang et al. (2026c) C. Jiang, Q. Zhao, L. Xin, J. Xie, P. Nakov, and Z. Xie Think with structured grounding: perceptual reinforcement learning for chart and visual-tabular understanding. arXiv preprint arXiv:2608.22429. Cited by: §2. Khan and Shang (2024) N. A. Khan and X. Shang Asd-gannet: a generative adversarial network-inspired deep learning approach for the classification of autism brain disorder. Brain Sciences 14 (8), p. 766. Cited by: §2. Khattak et al. (2023) M. U. Khattak, H. Rasheed, M. Maaz, S. Khan, and F. S. Khan Maple: multi-modal prompt learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, p. 19113–19122. Cited by: §1, §2. Kim et al. (2021) B. Kim, J. C. Ye, and J. Kim Learning dynamic graph representation of brain connectome with spatio-temporal attention. Advances in Neural Information Processing Systems 34, p. 4314–4327. Cited by: §2, §4.1. Li et al. (2021) X. Li, Y. Zhou, N. Dvornek, M. Zhang, S. Gao, J. Zhuang, D. Scheinost, L. H. Staib, P. Ventola, and J. S. Duncan Braingnn: interpretable brain graph neural network for fmri analysis. Medical image analysis 74, p. 102233. Cited by: §2, §4.1. Liu et al. (2024) X. Liu, M. R. Hasan, T. Gedeon, and M. Z. Hossain MADE-for-asd: a multi-atlas deep ensemble network for diagnosing autism spectrum disorder. Computers in Biology and Medicine 182, p. 109083. Cited by: §2. Lord (2019) C. Lord Recognising the heterogeneity of autism. The Lancet Psychiatry 6 (7), p. 551–552. Cited by: §1. Michelson et al. (2022) K. N. Michelson, J. G. Adams, and J. M. Faber Navigating clinical and business ethics while sharing patient data. Jama 327 (11), p. 1025–1026. Cited by: §1. Monajatipoor et al. (2024) M. Monajatipoor, Z. Dou, A. Chien, N. Peng, and K. Chang Medical vision-language pre-training for brain abnormalities. In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), p. 11159–11164. Cited by: §2. Phan et al. (2024) V. M. H. Phan, Y. Xie, Y. Qi, L. Liu, L. Liu, B. Zhang, Z. Liao, Q. Wu, M. To, and J. W. Verjans Decomposing disease descriptions for enhanced pathology detection: a multi-aspect vision-language pre-training framework. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, p. 11492–11501. Cited by: §2. Qu et al. (2026) C. Qu, Y. Zhong, X. Zhu, J. Li, C. Jiang, L. Jin, et al. Detect any ai-counterfeited text image. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, p. 35437–35450. Cited by: §2. Robins et al. (2014) D. L. Robins, K. Casagrande, M. Barton, C. A. Chen, T. Dumont-Mathieu, and D. Fein Validation of the modified checklist for autism in toddlers, revised with follow-up (m-chat-r/f). Pediatrics 133 (1), p. 37–45. Cited by: §1. Salari et al. (2022) N. Salari, S. Rasoulpoor, S. Rasoulpoor, S. Shohaimi, S. Jafarpour, N. Abdoli, B. Khaledi-Paveh, and M. Mohammadi The global prevalence of autism spectrum disorder: a comprehensive systematic review and meta-analysis. Italian journal of pediatrics 48 (1), p. 112. Cited by: §1. Sandbank et al. (2023) M. Sandbank, K. Bottema-Beutel, S. C. LaPoint, J. I. Feldman, D. J. Barrett, N. Caldwell, K. Dunham, J. Crank, S. Albarran, and T. Woynaroski Autism intervention meta-analysis of early childhood studies (project aim): updated systematic review and secondary analysis. Bmj 383. Cited by: §1. Santomauro et al. (2025) D. F. Santomauro, H. E. Erskine, A. M. M. Herrera, P. A. Miller, J. Shadid, H. Hagins, I. Y. Addo, Q. E. S. Adnani, B. O. Ahinkorah, A. Ahmed, et al. The global epidemiology and health burden of the autism spectrum: findings from the global burden of disease study 2021. The Lancet Psychiatry 12 (2), p. 111–121. Cited by: §1. Schielen et al. (2024) S. J. Schielen, J. Pilmeyer, A. P. Aldenkamp, and S. Zinger The diagnosis of asd with mri: a systematic review and meta-analysis. Translational psychiatry 14 (1), p. 318. Cited by: §2. She et al. (2021) J. She, Y. Hu, H. Shi, J. Wang, Q. Shen, and T. Mei Dive into ambiguity: latent distribution mining and pairwise uncertainty estimation for facial expression recognition. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, p. 6248–6257. Cited by: §2, §4.1. Shen et al. (2017) X. Shen, E. S. Finn, D. Scheinost, M. D. Rosenberg, M. M. Chun, X. Papademetris, and R. T. Constable Using connectome-based predictive modeling to predict individual behavior from brain connectivity. nature protocols 12 (3), p. 506–518. Cited by: §2, §4.1. Shorten and Khoshgoftaar (2019) C. Shorten and T. M. Khoshgoftaar A survey on image data augmentation for deep learning. Journal of big data 6 (1), p. 1–48. Cited by: §2. Team (2026) G. Team Venus-defakerone: unified fake image detection & localization. arXiv preprint arXiv:2605.14091. Cited by: §2. Uddin et al. (2024) M. Z. Uddin, M. A. Shahriar, M. N. Mahamood, F. Alnajjar, M. I. Pramanik, and M. A. R. Ahad Deep learning with image-based autism spectrum disorder analysis: a systematic review. Engineering Applications of Artificial Intelligence 127, p. 107185. Cited by: §1, §2. Wang et al. (2022) F. Wang, Y. Zhou, S. Wang, V. Vardhanabhuti, and L. Yu Multi-granularity cross-modal alignment for generalized medical visual representation learning. Advances in neural information processing systems 35, p. 33536–33549. Cited by: §2. Wang et al. (2023) H. Wang, C. Ma, J. Zhang, Y. Zhang, J. Avery, L. Hull, and G. Carneiro Learnable cross-modal knowledge distillation for multi-modal learning with missing modality. In International Conference on Medical Image Computing and Computer-Assisted Intervention, p. 216–226. Cited by: §2. Wang et al. (2020a) K. Wang, X. Peng, J. Yang, S. Lu, and Y. Qiao Suppressing uncertainties for large-scale facial expression recognition. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, p. 6897–6906. Cited by: §2, §4.1. Wang et al. (2020b) K. Wang, X. Peng, J. Yang, D. Meng, and Y. Qiao Region attention networks for pose and occlusion robust facial expression recognition. IEEE Transactions on Image Processing 29, p. 4057–4069. Cited by: §2, §4.1. Wieckowski et al. (2023) A. T. Wieckowski, L. N. Williams, J. Rando, K. Lyall, and D. L. Robins Sensitivity and specificity of the modified checklist for autism in toddlers (original and revised) a systematic review and meta-analysis. JAMA pediatrics 177 (4), p. 373–383. Cited by: §1. Wu et al. (2023) C. Wu, X. Zhang, Y. Zhang, Y. Wang, and W. Xie Medklip: medical knowledge enhanced language-image pre-training for x-ray diagnosis. In Proceedings of the IEEE/CVF international conference on computer vision, p. 21372–21383. Cited by: §2. Wu et al. (2024) R. Wu, H. Wang, H. Chen, and G. Carneiro Deep multimodal learning with missing modality: a survey. arXiv preprint arXiv:2409.07825. Cited by: §1, §2. Zeng et al. (2025a) F. Zeng, Z. Cheng, F. Zhu, H. Wei, and X. Zhang Local-prompt: extensible local prompts for few-shot out-of-distribution detection. In The Thirteenth International Conference on Learning Representations, Cited by: §2. Zeng et al. (2025b) F. Zeng, F. Zhu, H. Guo, X. Zhang, and C. Liu Modalprompt: towards efficient multimodal continual instruction tuning with dual-modality guided prompt. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, p. 12137–12152. Cited by: §1. Zeng et al. (2021) Y. Zeng, X. Zhang, and H. Li Multi-grained vision language pre-training: aligning texts with visual concepts. arXiv preprint arXiv:2111.08276. Cited by: §2. Zhang et al. (2021) Y. Zhang, C. Wang, and W. Deng Relative uncertainty learning for facial expression recognition. Advances in Neural Information Processing Systems 34, p. 17616–17627. Cited by: §2, §4.1. Zhang et al. (2022) Y. Zhang, C. Wang, X. Ling, and W. Deng Learn from all: erasing attention consistency for noisy label facial expression recognition. In European conference on computer vision, p. 418–434. Cited by: §2, §4.1. Zhao et al. (2025a) H. Zhao, F. Zhu, H. Guo, M. Wang, R. Wang, G. Meng, and Z. Zhang Mllm-cl: continual learning for multimodal large language models. arXiv preprint arXiv:2506.05453. Cited by: §2. Zhao et al. (2025b) Z. Zhao, E. Chung, K. Chung, and C. H. Park AV-fos: a transformer-based audio-visual multi-modal interaction style recognition for children with autism based on the family observation schedule (fos-i). IEEE journal of biomedical and health informatics. Cited by: §2. Zhou et al. (2024) Y. Zhou, G. Jia, Y. Ren, Y. Ren, Z. Xiao, and Y. Wang Advancing asd identification with neuroimaging: a novel garl methodology integrating deep q-learning and generative adversarial networks. BMC Medical Imaging 24 (1), p. 186. Cited by: §2.