CBAM-Enhanced Transfer Learning for Imbalanced Skin Lesion Classification
Students & Supervisors
Student Authors
Supervisors
Abstract
Skin cancer is one of the deadliest and rapidly increasing form of cancer worldwide. Early detection of this type cancer can contribute to decrease the mortality rate by skin disease cancer. An automation to the detection of the skin cancer can make medical procedure faster. Real world medical dataset such as HAM10000 which is one of the largest annotated datasets available, has sever class imbalance issue. This research highlights the problem of accurate multi-class skin lesion classificationa. By Conducting a comparative evaluation among two deep learning architectures with the merge of attention enhanced framework gives a better insight of this complexity. By integrating Convolutional Block Attention Module (CBAM) with pretrained moel ResNet50 and EfficientNet-B6 the imbalance problem was handled. Informative and channel-wise feature extraction has captured important lesion regions. Focal loss helps to mitigate the class imbalance and also class-aware weighting is applied for better class weights. As a result the ResNet50 model has perfomed good and achieved accuracy of 87%. But the most impressive result is showcased on EfficientNet-B6. This model achieved about 90% accuracy. Between the 2 models comparison EfficientNet-B6 has performed superior. As a result EfficientNet B6 + CBAM outperforming ResNet50 + CBAM which validates the effectivenes of attention mechanisms in imbalanced skin lesion classification. The main goal is to highlight the efficiency of the CBAM module with the pretrained model to enhance performance and improve detection process of the models.
Keywords
Publication Details
- Type of Publication:
- Conference Name: 2026 IEEE 2nd International Conference on Quantum Photonics, Artificial Intelligence, and Networking (QPAIN)
- Date of Conference: 16/04/2026 - 16/04/2026
- Venue: Chittagong, Bangladesh