Deep Learning-Assisted Early Detection of Skin Cancer from Dermoscopic Images in Underserved Clinical Settings

Area of Science:

• Dermatology and Artificial Intelligence (AI)

Background:

• Skin cancer, often caused by UV exposure, has high mortality rates due to late detection, particularly in regions with limited dermatological access.
• Early skin cancer detection is critical for improving patient outcomes and reducing mortality.
• Sociocultural factors and unequal distribution of healthcare exacerbate diagnostic challenges in many areas.

Purpose of the Study:

• To compare the performance of VGG16 and ResNet-50 deep learning models for automated skin cancer classification.
• To develop an AI-assisted approach for early detection of melanoma and non-melanoma skin cancers using dermoscopic images.
• To address challenges in skin cancer diagnosis in underserved regions through advanced image analysis.

Main Methods:

• Gathered a dataset of 1200 dermoscopic images from two clinics in Himachal Pradesh.
• Implemented a pipeline involving preprocessing, lesion segmentation, and classification using VGG16 and ResNet-50 models.
• Trained and evaluated the models on the annotated dermoscopic image dataset.

Main Results:

• ResNet-50 achieved 93% accuracy and 96% AUC-ROC, outperforming VGG16 (89% accuracy, 94% AUC-ROC).
• The study highlights the significant impact of model selection and preprocessing on diagnostic performance.
• Demonstrated the potential of AI in classifying dermoscopic images for skin cancer detection.

Conclusions:

• ResNet-50 shows superior performance for automated skin cancer classification from dermoscopic images.
• AI-assisted tools, with careful model selection and preprocessing, can significantly enhance early skin cancer detection.
• Future work includes ensemble methods, multi-class classification, and clinical validation for AI diagnostic tool implementation.