Dermatoscopic Image Analysis using Fine-Tuned ResNet50: Classifying Benign & Malignant Skin Lesions

Dermatological examinations have to rapidly detect skin lesions to improve patient outcomes and care. A refined ResNet50 transfer learning model classifies skin lesions as benign or malignant (cancers) in this study. A properly curated dataset of 3297 accurately labelled dermatoscopic pictures is used to test deep learning methods in dermatology. The dataset has three parts: 64% for training, 16% for validation, and 20% for testing. Partitioning facilitated model training, fine-tuning, and evaluation. The ResNet50 model was extensively trained on 2109 images. A separate selection of 528 pictures was used to evaluate and verify the model. This strategy maximised performance and minimised overfitting. A selection of 660 images that accurately showed a variety of real-life settings was used to evaluate the model. ResNet50, noted for its capacity to extract detailed visual information, was fine-tuned to reliably categorise skin lesions. The model accurately classified benign and malignant tumours with 93.33% accuracy. High accuracy indicates the model's great and dependable performance. Performance was measured beyond precision. These measurements included using accuracy and loss charts during training and a confusion matrix during testing. The thorough evaluation enabled the calculation of various performance metrics, demonstrating the model's dependability and efficacy in dermatologists' diagnostic processes. This paper examines the dermatology applications of deep learning (DL) models, notably the fine-tuned ResNet50 model. The model's 93.33% accuracy shows its ability to improve diagnostic accuracy. This could greatly improve clinical practises and benefit patients and healthcare systems.