Early Non-Invasive Skin Cancer Diagnosis using ResNet-50 Model

Skin cancer is one of the most life-threatening and widespread types of cancer globally, which necessitates early detection to enhance patient care. Traditional methods of detection, including biopsies and histopathological testing, are usually invasive, time-consuming, and painful to patients. To address these issues, this research suggests a non-invasive deep learning method using ResNet-50 for early detection of skin cancer. ResNet-50, a deep convolutional neural networks (CNN), is well suited for classifying medical images because of its residual connections that successfully solve the vanishing gradient problem in deep networks, allowing better feature extraction and classification performance. The model is trained with publicly available datasets of skin cancers, such as the ISIC Skin Cancer MNIST: HAM10000 and the Skin Cancer 9 Classes (ISIC) dataset. These data consist of thousands of labeled dermatoscopic images of benign and malignant skin lesions and include a diverse and large collection of training samples. A strong data preprocessing pipeline is used to optimize model performance, such as normalization, resizing of images, and data augmentation processes like rotation, flipping, zooming, and brightness modification. These preprocessing procedures enhance the model's generalization capability and robustness against variability in real-world skin lesion images. In the training process, the pre-trained ResNet-50 model is fine-tuned by adjusting its last fully connected layer to support both binary classification (benign or malignant) and multi-class classification of skin lesions.