Comprehensive Summary
This paper presents a novel deep learning framework designed to classify medical images, focusing on feature extraction and classification across five cancer types: skin, breast, lung, oral, and stomach cancer. The researchers utilized five publicly available datasets (INBreast, KVASIR, ISIC2018, Lung Cancer, and Oral Cancer) and applied data augmentation to address dataset imbalance. A self-attention-based convolutional neural network (CNN) was developed, incorporating an inverted residual with multiscale weight layers to optimize feature extraction and selection. The proposed architecture achieved high accuracy rates across the datasets: 98.6% for INBreast, 95.3% for KVASIR, 94.3% for ISIC2018, 95.0% for Lung Cancer, and 98.8% for Oral Cancer. Precision rates improved significantly post-data augmentation. The study also compared the new architecture's performance against four pre-trained CNN models (AlexNet, GoogleNet, ResNet50, and Densenet201), demonstrating superior accuracy by 4-5%. The findings underscore the importance of data augmentation and advanced CNN architectures in enhancing classification accuracy.
Outcomes and Implications
The research holds significant clinical implications, particularly in the early detection and classification of various cancer types through medical imaging. By improving feature extraction and classification accuracy, the proposed framework can potentially enhance diagnostic precision and treatment planning in oncology. The ability to accurately classify images of skin, breast, lung, stomach, and oral cancers could lead to earlier interventions and better patient outcomes. Moreover, the study addresses common challenges in medical imaging, such as imbalanced datasets and redundant information, offering a robust solution that can be integrated into clinical workflows. The advancement in CNN architectures as demonstrated in this study could pave the way for more reliable and efficient diagnostic tools in the medical field.