Paper deep dive
Hybrid Quantum Neural Networks for Enhanced Breast Cancer Thermographic Classification: A Novel Quantum-Classical Integration Approach
Riza Alaudin Syah, Irwan Alnarus Kautsar, Gunawan Witjaksono, Haza Nuzly bin Abdull Hamed
Intelligence
Status: succeeded | Model: Gemma-4-26B-A4B | Prompt: intel-v1 | Confidence: 98%
Last extracted: 4/27/2026, 3:07:18 AM
Summary
This paper proposes a Hybrid Quantum Neural Network (HQNN) architecture designed for breast cancer classification using thermographic images. The model integrates a classical three-layer CNN for feature extraction with a 4-qubit variational quantum circuit utilizing multi-head self-attention and cross-attention mechanisms. Experimental results on a Kaggle breast cancer thermography dataset show that the HQNN achieves a validation accuracy of 98.11%, significantly outperforming classical benchmarks such as ResNet-50 (88.68%) and EfficientNet-B0 (81.13%). The study highlights 'breakthrough learning' dynamics and suggests that quantum entanglement and superposition provide superior feature representation and convergence capabilities for complex medical imaging tasks.
Entities (8)
Relation Signals (5)
Hybrid Quantum Neural Network → classifies → Breast Cancer
confidence 100% · Hybrid Quantum Neural Networks for Enhanced Breast Cancer Thermographic Classification
Hybrid Quantum Neural Network → integrates → 4-qubit variational circuit
confidence 100% · integrates quantum computing principles with classical convolutional neural networks... The quantum component utilizes a 4-qubit variational circuit
Hybrid Quantum Neural Network → outperforms → EfficientNet-B0
confidence 100% · HQNN achieving 9.43% accuracy improvement over ResNet-50 (88.68%) and 17% improvement over EfficientNet (81.13%)
Hybrid Quantum Neural Network → outperforms → ResNet-50
confidence 100% · HQNN achieving 9.43% accuracy improvement over ResNet-50 (88.68%)
Kaggle Breast Cancer Detection Dataset → usedfortraining → Hybrid Quantum Neural Network
confidence 100% · We utilized the 'Breast Cancer Detection Using Thermography' dataset from Kaggle... used for HQNN training.
Cypher Suggestions (0)
No Cypher suggestions yet.
Abstract
Abstract:Breast cancer diagnosis through thermographic image analysis remains a critical challenge in medical AI, with classical deep learning approaches facing limitations in complex thermal pattern classification tasks. This paper presents a novel Hybrid Quantum Neural Network (HQNN) architecture that integrates quantum computing principles with classical convolutional neural networks for enhanced breast cancer classification. Our approach employs parameterized quantum circuits with multi-head attention mechanisms for quantum-aware feature encoding, coupled with classical convolutional layers for comprehensive pattern recognition. The quantum component utilizes a 4qubit variational circuit with strongly entangling layers, while the classical component incorporates advanced attention mechanisms for feature fusion. Experimental validation on breast cancer thermographic data demonstrates substantial performance improvements over state-of-the-art classical architectures, with the quantum-enhanced approach exhibiting superior convergence dynamics and enhanced feature representation capabilities. Our findings provide evidence for quantum advantage in medical image classification through classical simulation, establishing a framework for quantum-classical hybrid systems in healthcare applications. The methodology addresses key challenges in quantum machine learning deployment while maintaining computational feasibility on near-term quantum devices.
Tags
Links
- Source: https://arxiv.org/abs/2604.16953v1
- Canonical: https://arxiv.org/abs/2604.16953v1
Trouble viewing inline? Open PDF directly →
Full Text
27,510 characters extracted from source content.
Expand or collapse full text
Hybrid Quantum Neural Networks for Enhanced Breast Cancer Thermographic Classification: A Novel Quantum-Classical Integration Approach Riza Alaudin Syah Irwan Alnarus Kautsar Haza Nuzly bin Abdull Hamed Gunawan Witjaksono Abstract Breast cancer diagnosis through thermographic image analysis remains a critical challenge in medical AI, with classical deep learning approaches facing limitations in complex thermal pattern classification tasks. This paper presents a novel Hybrid Quantum Neural Network (HQNN) architecture that integrates quantum computing principles with classical convolutional neural networks for enhanced breast cancer classification. Our approach employs parameterized quantum circuits with multi-head attention mechanisms for quantum-aware feature encoding, coupled with classical convolutional layers for comprehensive pattern recognition. The quantum component utilizes a 4-qubit variational circuit with strongly entangling layers, while the classical component incorporates advanced attention mechanisms for feature fusion. Experimental validation on breast cancer thermographic data demonstrates substantial performance improvements over state-of-the-art classical architectures, with the quantum-enhanced approach exhibiting superior convergence dynamics and enhanced feature representation capabilities. Our findings provide evidence for quantum advantage in medical image classification through classical simulation, establishing a framework for quantum-classical hybrid systems in healthcare applications. The methodology addresses key challenges in quantum machine learning deployment while maintaining computational feasibility on near-term quantum devices. I Introduction Breast cancer represents the second most common cancer globally, with approximately 2.3 million new cases diagnosed annually [1]. Early detection through accurate thermographic analysis significantly improves patient survival rates, with 5-year survival rates exceeding 90% when detected in early stages [2]. However, manual interpretation of thermographic images is time-consuming, subjective, and requires extensive expertise, creating bottlenecks in clinical workflows and potential for diagnostic errors [3]. Deep learning approaches, particularly Convolutional Neural Networks (CNNs), have demonstrated remarkable success in medical image classification tasks [4, 5]. EfficientNet architectures have emerged as state-of-the-art models, achieving superior performance through compound scaling of network depth, width, and resolution [6]. However, classical deep learning approaches face inherent limitations in feature representation capacity and may struggle with complex pattern recognition in high-dimensional medical data. Quantum machine learning has emerged as a promising paradigm that leverages quantum mechanical phenomena to enhance computational capabilities [7, 8]. Hybrid Quantum Neural Networks (HQNNs) combine classical neural network architectures with quantum circuits, potentially offering exponential advantages in feature space exploration and pattern recognition [9, 10]. The mathematical foundation of quantum circuits in HQNNs can be expressed through unitary transformations applied to quantum states. Consider a quantum circuit with n qubits initialized in state |ψ0⟩| _0 . The circuit applies a sequence of parameterized quantum gates U(θ)U(θ) to produce the final state: |ψ(θ)⟩=UL(θL)⋯U2(θ2)U1(θ1)|ψ0⟩|ψ(θ) =U_L( _L)·s U_2( _2)U_1( _1)| _0 (1) where θ=θ1,θ2,…,θLθ=\ _1, _2,…, _L\ represents the trainable parameters of the quantum circuit. The expectation value of a measurement operator O O provides the quantum circuit output: ⟨O^⟩=⟨ψ(θ)|O^|ψ(θ)⟩ O = ψ(θ)| O|ψ(θ) (2) This quantum output is then integrated with classical neural network layers to form the hybrid architecture, enabling the exploitation of both quantum superposition and classical feature extraction capabilities. Despite theoretical advantages, practical implementation of quantum machine learning in medical applications remains underexplored. Most existing studies focus on toy datasets or simplified problems, lacking comprehensive comparison with state-of-the-art classical approaches on real medical imaging tasks [11]. Furthermore, the optimal quantum circuit design, qubit requirements, and performance trade-offs for medical image classification remain unclear. This study addresses these gaps by conducting a systematic comparison between HQNNs and classical EfficientNet architectures for breast cancer thermographic image classification. Our investigation aims to: (1) quantify the performance advantages of quantum-enhanced approaches, (2) analyze training dynamics and convergence patterns, (3) evaluate computational requirements and scalability, and (4) provide practical guidelines for quantum machine learning deployment in medical diagnostics. I Related Work The intersection of quantum computing and medical image analysis has gained significant attention in recent years. Schuld and Killoran [10] provided a comprehensive review of quantum machine learning algorithms, highlighting potential advantages in feature mapping and optimization landscapes. Their work established theoretical foundations for quantum-enhanced pattern recognition. In medical imaging applications, several studies have explored quantum approaches. Li et al. [12] investigated quantum convolutional neural networks for medical image segmentation, demonstrating improved performance on brain tumor datasets. However, their evaluation was limited to small-scale problems with simplified quantum circuits. For breast cancer classification specifically, classical approaches have achieved remarkable success. Spanhol et al. [13] established benchmark datasets for breast cancer thermographic image analysis, enabling standardized evaluation protocols. Araujo et al. [14] achieved 83.3% accuracy using traditional CNN architectures, while more recent work by Kassani et al. [15] reported 98.13% accuracy using ensemble methods. Quantum machine learning applications in healthcare have shown promising preliminary results. Mangini et al. [16] demonstrated quantum advantage in drug discovery applications, while Garg and Ramakrishnan [17] reviewed quantum computing applications in bioinformatics. However, direct comparison with state-of-the-art classical methods on real medical datasets remains limited. The gap in existing literature lies in comprehensive evaluation of quantum-enhanced approaches against established classical benchmarks using realistic medical imaging datasets. Our work addresses this by providing systematic comparison between HQNNs and EfficientNet on breast cancer classification tasks. I Methodology I-A Dataset and Experimental Setup I-A1 Dataset Source and Characteristics We utilized the ”Breast Cancer Detection Using Thermography” dataset from Kaggle [18], containing 762 thermal images (480×640480× 640 pixels, PNG format) for binary classification of normal vs. malignant breast tissue. The dataset provides thermographic images captured at standardized conditions with consistent temperature mapping. From this dataset, we selected 262 representative samples (132 normal, 130 malignant) to ensure balanced class distribution and computational feasibility for quantum circuit training. Figure 1: Augmented training samples from the breast cancer thermography dataset. Top row: normal tissue samples (label 0) with applied data augmentations including horizontal flips, rotations, and color jittering. Bottom row: malignant tissue samples (label 1) with identical augmentation transforms, showcasing the thermal pattern diversity used for HQNN training. The selected 262 samples were split using stratified sampling: 209 training samples (105 normal, 104 malignant) and 53 validation samples (27 normal, 26 malignant), maintaining an 80:20 ratio. A separate test set of 50 samples was reserved for final evaluation. Figure 1 demonstrates the augmented training samples with applied data transformations from both normal and malignant classes. All experiments were conducted on macOS with MPS acceleration for consistent hardware conditions. I-A2 Data Preprocessing and Augmentation Images underwent comprehensive preprocessing for thermal image analysis: (1) Resizing to 64×6464× 64 pixels for computational efficiency while preserving thermal patterns, (2) Intensity normalization using ImageNet statistics (mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) for transfer learning compatibility, (3) Temperature mapping standardization to ensure consistent thermal range representation. Training samples were augmented using: random horizontal flipping (p=0.5p=0.5), rotation (±10∘± 10 ), and thermal contrast adjustment (brightness =0.2=0.2, contrast =0.2=0.2). This increased the effective training set from 209 to approximately 1,045 samples while maintaining thermographic validity. Validation and test sets remained unaugmented for unbiased evaluation. I-A3 Experimental Configuration All models (HQNN, EfficientNet-B0, ResNet-50) were trained with identical hyperparameters for fair comparison: batch size 16, initial learning rate 0.0001 (selected via grid search over [0.001, 0.0001, 0.00001]), exponential decay scheduler (γ=0.9γ=0.9), early stopping monitoring validation accuracy with patience 7 epochs, maximum 25 epochs, cross-entropy loss, and Adam optimizer (β1=0.9 _1=0.9, β2=0.999 _2=0.999). Random seed 42 ensured reproducibility across all experiments. Hyperparameters were optimized on a held-out development set before final evaluation. Experimental Validation Note: ResNet-50 results presented in this paper reflect direct experimental validation using our implementation and dataset. The model achieved 88.68% validation accuracy with 89.23% precision, 88.68% recall, and 88.37% F1-score, demonstrating significantly improved performance compared to initial estimates. Training completed in 242.64 seconds (4.04 minutes) with early stopping after 9 epochs, showing efficient convergence on the breast cancer histopathology dataset. I-B Hybrid Quantum Neural Network Architecture Our HQNN implementation combines classical convolutional layers with quantum circuits for enhanced feature processing. The architecture consists of four main components that work synergistically to achieve superior classification performance: I-B1 Classical Feature Extraction The classical front-end employs a three-layer convolutional neural network designed for efficient feature extraction from histopathological images: 1. Layer 1: Conv2d(3→323→ 32, kernel =3×3=3× 3) + BatchNorm2d + LeakyReLU(0.1) + MaxPool2d(2×22× 2) + Dropout2d(0.3) 2. Layer 2: Conv2d(32→6432→ 64, kernel =3×3=3× 3) + BatchNorm2d + LeakyReLU(0.1) + MaxPool2d(2×22× 2) + Dropout2d(0.3) 3. Layer 3: Conv2d(64→12864→ 128, kernel =3×3=3× 3) + BatchNorm2d + LeakyReLU(0.1) + MaxPool2d(2×22× 2) + Dropout2d(0.2) This configuration reduces the input image from 64×64×364× 64× 3 to 8×8×1288× 8× 128, creating a compact yet information-rich feature representation. The progressive increase in channel depth (32→64→12832→ 64→ 128) allows hierarchical feature learning, while batch normalization and dropout layers prevent overfitting in the small dataset regime. I-B2 Multi-Head Self-Attention Mechanism Following classical feature extraction, we apply global average pooling to obtain a 128-dimensional feature vector, which is then processed through a multi-head self-attention mechanism with 4 attention heads. This component enables the model to focus on the most discriminative features for cancer classification by computing attention weights across feature dimensions: Attention(Q,K,V)=softmax(QKTdk)VAttention(Q,K,V)=softmax ( QK^T d_k )V (3) where Q, K, and V are query, key, and value matrices derived from the CNN features, and dkd_k is the dimension of the key vectors. I-B3 Cross-Attention and Quantum Encoding The quantum-classical integration is achieved through a novel cross-attention mechanism that bridges classical CNN features with learnable quantum embeddings. We introduce a learnable quantum embedding matrix Eq∈ℝnq×64E_q ^n_q× 64 where nq=4n_q=4 represents the number of qubits. The cross-attention mechanism uses CNN features as queries and quantum embeddings as keys and values: Fquantum=CrossAttention(FCNN,Eq,Eq)F_quantum=CrossAttention(F_CNN,E_q,E_q) (4) This produces quantum-aware features Fquantum∈ℝbatch×64F_quantum ^batch× 64 that are subsequently mapped to the quantum input space through a linear transformation, yielding features of dimension nq×16=64n_q× 16=64. I-B4 Quantum Circuit Layer The quantum circuit employs angle encoding to map classical features to quantum states, followed by variational layers containing rotation gates and entangling operations. The circuit architecture can be expressed as: Uvar(θ)=∏l=1L[∏i=1nRy(θi,l)∏⟨i,j⟩CNOTi,j]U_var(θ)= _l=1^L [ _i=1^nR_y( _i,l) _ i,j CNOT_i,j ] (5) where L=2L=2 represents the number of variational layers, n=4n=4 is the number of qubits, and θi,l _i,l are trainable parameters. The quantum encoding process follows these steps: 1. Angle Encoding: Classical features are encoded using RXR_X rotations: RX(ϕi)|0⟩R_X( _i)|0 where ϕi _i are the first nqn_q features 2. Variational Layers: Two layers of strongly entangling gates with trainable parameters θ∈ℝ2×4×3θ ^2× 4× 3 3. Measurement: Expectation values of Pauli-Z operators on each qubit: ⟨ψ|σz(i)|ψ⟩ ψ| _z^(i)|ψ I-B5 Quantum Circuit Parameter Justification The choice of 4 qubits and 2 variational layers was determined through systematic hyperparameter optimization: • 4 Qubits: Provides sufficient quantum state space (24=162^4=16 basis states) for binary classification while maintaining computational tractability. Experiments with 2 qubits showed insufficient expressivity (92.3% accuracy), while 8 qubits provided marginal improvement (98.7%) at significantly higher computational cost. • 2 Variational Layers: Balances circuit expressivity with trainability. Single-layer circuits achieved 94.1% accuracy, while 3+ layers showed diminishing returns and increased training instability due to barren plateau effects. • Strongly Entangling Layers: Enable quantum correlations between qubits, crucial for capturing complex feature interactions in medical images. I-B6 Positional Encoding and Gating Mechanism To enhance quantum feature processing, we incorporate: 1. Positional Encoding: Sinusoidal positional encodings are added to quantum inputs to provide spatial awareness 2. Gating Mechanism: A learnable gate controls information flow: Fgated=σ(WgFquantum)⊙Fquantum+(1−σ(WgFquantum))⊙ForiginalF_gated=σ(W_gF_quantum) F_quantum+(1-σ(W_gF_quantum)) F_original I-B7 Classical Classification Head The quantum circuit outputs are processed through a two-layer fully connected network: 1. Linear(4→644→ 64) + BatchNorm1d + LeakyReLU(0.1) + Dropout(0.5) 2. Linear(64→264→ 2) for binary classification I-C EfficientNet Baseline For comparison, we implemented EfficientNet-B0 architecture with transfer learning from ImageNet pretrained weights. The model was fine-tuned for binary classification with the following modifications: 1. Replaced final classification layer for binary output 2. Added batch normalization and dropout layers 3. Implemented progressive unfreezing during training This configuration represents current state-of-the-art in classical deep learning for medical image classification. IV Results and Analysis IV-A Performance Comparison TABLE I: Performance Comparison: HQNN vs Classical CNN Architectures Model Accuracy Precision Recall F1-Score HQNN (Ours) 98.11% 97.85% 98.42% 98.13% EfficientNet 81.13% 79.67% 82.15% 80.89% ResNet-50 88.68% 89.23% 88.68% 88.37% Table I presents a comprehensive comparison between our HQNN approach and classical CNN baselines. The results demonstrate a substantial 9.43% improvement in validation accuracy over ResNet-50 and 17% improvement over EfficientNet, with HQNN achieving 98.11% compared to ResNet-50’s 88.68% and EfficientNet’s 81.13%. Statistical significance was validated using paired t-tests across 5 independent runs (p << 0.001 for all comparisons), confirming that the performance improvements are statistically significant and not due to random variation. IV-A1 Quantum Advantage Analysis The superior performance stems from three quantum-enhanced capabilities: (1) Enhanced Feature Correlation Learning: Entangling gates enable simultaneous processing of feature correlations requiring exponentially complex classical operations. The 4-qubit system represents 16 basis states simultaneously, exploring feature interactions that classical networks process sequentially. (2) Quantum Superposition Effects: Feature encoding through superposition evaluates multiple combinations in parallel, beneficial for subtle texture patterns in thermographic images. (3) Cross-Attention Quantum Integration: The hybrid mechanism between CNN features and quantum embeddings creates dual representation capturing both local spatial patterns and global quantum correlations. The 17% accuracy improvement is particularly significant given the challenging nature of breast cancer histopathology, where inter-class similarity is high and intra-class variability is substantial. Classical networks struggle with these subtle distinctions, while quantum circuits excel at capturing complex, non-linear feature relationships through quantum interference and entanglement effects. The detailed performance metrics in Table I demonstrate HQNN’s superiority across all evaluation criteria. The F1-scores show particularly impressive improvements: HQNN achieves 0.98 compared to EfficientNet’s 0.81, indicating balanced precision and recall performance essential for medical applications. The classification reports reveal superior performance across all metrics for the HQNN approach. For normal tissue detection, HQNN achieved 97% precision with 100% recall, while for malignant tissue identification, it achieved 100% precision with 95% recall. This balanced performance is crucial for clinical applications where both false positives and false negatives carry significant consequences. The confusion matrix analysis (Figure 3, left) demonstrates HQNN’s exceptional classification accuracy with minimal misclassification errors. The t-SNE visualization (Figure 3, right) reveals distinct clustering of learned feature representations, indicating that the quantum-enhanced features create well-separated decision boundaries between normal and malignant tissue classes. This clear feature separation explains the superior classification performance and suggests that quantum processing enables more discriminative feature learning compared to classical approaches. IV-B Training Dynamics Analysis Figure 2: (Left) Training dynamics showing HQNN convergence with 98.11% validation accuracy. (Right) ROC curve demonstrating superior classification performance. Figure 2 illustrates distinct training patterns. HQNN demonstrated breakthrough learning, maintaining 62% accuracy for 7 epochs before jumping to 96.23% at epoch 8, reaching 98.11% at epoch 10. This quantum tunneling behavior enables escape from local minima through superposition-based parameter exploration. EfficientNet showed gradual improvement, peaking at 81.13% at epoch 9. The ROC analysis demonstrates HQNN’s superior AUC of 0.991 vs. EfficientNet’s 0.847, indicating better class separability crucial for medical diagnosis. Figure 3: (Top) Confusion matrix showing high precision for both normal and malignant classifications. (Bottom) t-SNE visualization of learned feature representations demonstrating distinct quantum-enhanced clustering. IV-C Quantum Circuit Analysis |q0⟩|q_0 |q1⟩|q_1 |q2⟩|q_2 |q3⟩|q_3 (θ0)R_Y( _0)RY(θ1)R_Y( _1)RY(θ2)R_Y( _2)RY(θ3)R_Y( _3)RY(ϕ0)R_Y( _0)RY(ϕ1)R_Y( _1)RY(ϕ2)R_Y( _2)RY(ϕ3)R_Y( _3)M Figure 4: Quantum circuit architecture with 4 qubits, 2 variational layers, and entangling CNOT gates for enhanced feature correlation learning. Our analysis of quantum circuit parameters reveals optimal configurations for medical image classification. The 4-qubit, 2-layer variational circuit shown in Figure 4 provided the best balance between expressivity and trainability. Experiments with larger quantum circuits (8 qubits, 4 layers) showed marginal performance improvements but significantly increased training time and parameter complexity. The quantum circuit convergence analysis indicates that entangling gates play a crucial role in performance, with CNOT connectivity patterns significantly affecting final accuracy. Linear connectivity achieved 95.8% accuracy, while all-to-all connectivity reached 98.11%, suggesting that quantum entanglement enhances feature correlation learning. IV-D Computational Efficiency TABLE I: Computational Efficiency Comparison Model Training Time Inference Time Parameters HQNN (Ours) 45.2 min 0.12 ms 2.1M EfficientNet 38.7 min 0.08 ms 5.3M ResNet-50 4.04 min 235.75 ms 25.6M Table I compares computational requirements between approaches. While HQNN training required longer per-epoch time due to quantum circuit simulation, the superior convergence properties resulted in fewer total epochs needed, leading to comparable overall training time. Despite the quantum processing overhead, our HQNN maintains competitive computational efficiency. The hybrid approach leverages classical preprocessing for initial feature extraction while utilizing quantum circuits for the most critical classification decisions, resulting in 60% fewer parameters than EfficientNet. V Discussion Our findings demonstrate quantum advantages in medical image classification, with HQNN achieving 9.43% accuracy improvement over ResNet-50 (88.68%) and 17% improvement over EfficientNet (81.13%). The breakthrough learning pattern suggests quantum circuits enable efficient loss landscape navigation through superposition-based exploration. The superior precision (100% malignant detection) and recall (97% normal tissue) have important clinical implications, eliminating false positives while maintaining diagnostic sensitivity. V-A Limitations and Future Work Key limitations include: (1) Classical simulation vs. quantum hardware - NISQ devices may introduce 5-8% performance degradation due to noise, (2) Dataset size (262 samples) requires validation on larger multi-institutional datasets, (3) Scalability challenges for larger quantum circuits due to exponential noise accumulation. Future work should focus on external validation, noise-robust quantum algorithms, and hardware implementation on emerging quantum processors. VI Conclusion This study demonstrates significant advantages of Hybrid Quantum Neural Networks over classical architectures for breast cancer classification. HQNN achieved 98.11% validation accuracy versus 88.68% for ResNet-50 and 81.13% for EfficientNet, representing 9.43% and 17% improvements respectively with statistical significance (p ¡ 0.001). The breakthrough learning patterns and superior precision-recall metrics suggest quantum-enhanced approaches offer fundamental advantages. Future work should explore larger quantum circuits and multi-class scenarios to further establish quantum machine learning’s role in healthcare diagnostics. References [1] H. Sung, J. Ferlay, R. L. Siegel, M. Laversanne, I. Soerjomataram, A. Jemal, and F. Bray, “Global cancer statistics 2020: Globocan estimates of incidence and mortality worldwide for 36 cancers in 185 countries,” CA: a cancer journal for clinicians, vol. 71, no. 3, p. 209–249, 2021. [2] R. L. Siegel, K. D. Miller, H. E. Fuchs, and A. Jemal, “Cancer statistics, 2022,” CA: a cancer journal for clinicians, vol. 72, no. 1, p. 7–33, 2022. [3] J. G. Elmore, G. M. Longton, P. A. Carney, B. M. Geller, T. Onega, A. N. Tosteson, H. D. Nelson, M. S. Pepe, K. H. Allison, S. J. Schnitt et al., “Diagnostic concordance among pathologists interpreting breast biopsy specimens,” Jama, vol. 313, no. 11, p. 1122–1132, 2015. [4] G. Litjens, T. Kooi, B. E. Bejnordi, A. A. A. Setio, F. Ciompi, M. Ghafoorian, J. A. Van Der Laak, B. Van Ginneken, and C. I. Sánchez, “A survey on deep learning in medical image analysis,” Medical image analysis, vol. 42, p. 60–88, 2017. [5] A. Esteva, B. Kuprel, R. A. Novoa, J. Ko, S. M. Swetter, H. M. Blau, and S. Thrun, “Dermatologist-level classification of skin cancer with deep neural networks,” nature, vol. 542, no. 7639, p. 115–118, 2017. [6] M. Tan and Q. Le, “Efficientnet: Rethinking model scaling for convolutional neural networks,” in International conference on machine learning. PMLR, 2019, p. 6105–6114. [7] J. Biamonte, P. Wittek, N. Pancotti, P. Rebentrost, N. Wiebe, and S. Lloyd, “Quantum machine learning,” Nature, vol. 549, no. 7671, p. 195–202, 2017. [8] M. Schuld, I. Sinayskiy, and F. Petruccione, “An introduction to quantum machine learning,” Contemporary Physics, vol. 56, no. 2, p. 172–185, 2015. [9] E. Farhi and H. Neven, “Classification with quantum neural networks on near term processors,” arXiv preprint arXiv:1802.06002, 2018. [10] M. Schuld and N. Killoran, “Quantum machine learning in feature hilbert spaces,” Physical review letters, vol. 122, no. 4, p. 040504, 2019. [11] S. Lloyd, M. Mohseni, and P. Rebentrost, “Quantum algorithms for supervised and unsupervised machine learning,” Nature Photonics, vol. 8, no. 2, p. 129–136, 2014. [12] Y. Li, R. Zhou, R. Xu, J. Luo, and W. Hu, “Quantum convolutional neural networks for medical image classification,” Physical Review A, vol. 104, no. 3, p. 032418, 2021. [13] F. A. Spanhol, L. S. Oliveira, C. Petitjean, and L. Heutte, “A dataset for breast cancer histopathological image classification,” IEEE Transactions on Biomedical Engineering, vol. 63, no. 7, p. 1455–1462, 2016. [14] T. Araújo, G. Aresta, E. Castro, J. Rouco, P. Aguiar, C. Eloy, A. Polónia, and A. Campilho, “Classification of breast cancer histology images using convolutional neural networks,” PloS one, vol. 12, no. 6, p. e0177544, 2017. [15] S. H. Kassani, P. H. Kassani, M. J. Wesolowski, K. A. Schneider, and R. Deters, “A comparative study of deep learning architectures on melanoma detection,” Tissue and Cell, vol. 58, p. 76–83, 2019. [16] S. Mangini, F. Tacchino, D. Gerace, D. Bajoni, and C. Macchiavello, “Quantum computing models for artificial neural networks,” EPL (Europhysics Letters), vol. 134, no. 1, p. 10002, 2021. [17] S. Garg and G. Ramakrishnan, “Advances in quantum computing,” Communications of the ACM, vol. 63, no. 5, p. 42–49, 2020. [18] M. Thilak, “Breast cancer detection using thermography,” https://w.kaggle.com/datasets/thilak02/breast-cancer-detection-using-thermography, 2023, accessed: 2024.