Hierarchical MultiView Feature Aggregation and Explainable AI for Melanoma and NonMelanoma Classification

Melanoma, a severe form of skin cancer, is known for its high mortality rate if not detected early. Accurate and timely classification of skin lesions is critical for effective diagnosis and treatment. This project presents a Hierarchical Multi-View Feature Aggregation approach coupled with Explainable AI for classifying Melanoma and Non-Melanoma skin lesions. The classification model leverages Convolutional Neural Networks (CNN), MobileNet, and Grad-CAM (Gradient-weighted Class Activation Mapping) for interpretability, providing a holistic view of skin images while ensuring accurate and explainable predictions.

We utilize a combination of CNN and MobileNet to extract deep features from skin lesion images, where CNN captures hierarchical spatial features while MobileNet ensures efficiency and speed without compromising performance. To enhance the model’s interpretability, we incorporate Grad-CAM, a visual explanation technique that highlights the important regions of an image that contribute to the model’s decision. This ensures that the model’s predictions are not only accurate but also interpretable, allowing clinicians to understand the areas in the image that influence classification decisions.

The model is trained on a dataset from Kaggle, which includes a diverse set of labeled skin lesion images representing both melanoma and non-melanoma classes. The integration of multi-view feature aggregation allows the model to process images from different perspectives, improving robustness and classification accuracy. By combining deep learning with explainable AI, this approach not only provides reliable classification but also addresses the critical need for transparency in medical AI applications.

Our results show promising accuracy in classifying melanoma vs. non-melanoma lesions, and the integration of Grad-CAM ensures that each prediction is accompanied by an intuitive and meaningful explanation. This method provides a scalable solution for automated melanoma detection while ensuring trust in AI-based medical diagnostics.

Keywords: Melanoma, Non-Melanoma, CNN, MobileNet, Grad-CAM, Explainable AI, Skin Lesion Classification, Deep Learning, Hierarchical Multi-View Feature Aggregation.