Skin cancer image classification using hybrid quantum deep learning model with BiLSTM and MobileNetV2

Abstract Skin cancer image classification is known to be extremely complex due to the subtle visual differences between benign and malignant lesions. In this study, we propose a novel hybrid model that leverages the hierarchical feature extraction capabilities of the hybrid quantum convolutional neural network (HQCNN), the temporal dynamics captured by the bidirectional long short-term memory neural networks (BiLSTM) model, and the efficient feature extraction capabilities of MobileNetV2. We evaluated the proposed model on a clinically relevant skin cancer dataset, using images resized to 32 × 32 and 128 × 128 pixels to investigate the impact of resolution on classification performance. The HQCNN model augmented with BiLSTM and MobileNetV2 achieved a training accuracy of 97.7% and a test accuracy of 89.3% on 128 × 128-pixel color images, along with an F1 score of 89.81% and a recall of 94.33% for malignant cases, confirming clinical reliability and strong sensitivity in cancer detection. These results demonstrate robust feature extraction, improved contextual learning, and strong generalization for complex medical image classification tasks.