DeepSkin A Deep Learning Approach for Skin Cancer Classification

Aim:

            To design and develop Deep Learning Model that enable early and accurate diagnosis of skin lesions, improving the chances of successful treatment.

Synopsis:

            Given the rising prevalence of skin cancer and the significance for early detection, it is crucial to develop an effective method to automatically classify the skin cancer. As the largest organ of the human body, the skin shoulders the responsibility of protecting other human systems, which increases its vulnerability to disease. Melanoma was the most common cancer in both men and women with approximately 30000 new cases diagnosed in 2018. In addition to the melanoma, two other major skin cancer diseases, basal cell carcinoma (BCC) and squamous cell carcinoma (SCC) also had a relatively high incidence with over 1 million cases in 2018. This document discusses the high incidence of melanoma, basal cell carcinoma (BCC) and squamous cell carcinoma (SCC) and highlights the limitation of traditional machine learning methods in addressing diagnostic demands. It introduces the transition to deep learning algorithms for skin cancer classification, acknowledging their advantages in handling large scale datasets and aiding clinicians. However, it also addresses the challenges faced by deep learning, such as data imbalance, computational costs and issues with robustness and generalization. The review criticizes existing literatures for not thoroughly analyzing practical constraints in clinical setting, including data imbalance, cross-domain adaptability, model robustness and efficiency. The paper aims to provide a comprehensive summary of frontier challenges in skin cancer classification, offering insights into the advancements and limitation of deep learning.

Proposed System:

            The proposed method focus on skin cancer classification using the MobileNetV2 deep learning architecture. The proposed method begins with loading and visualizing segmented skin disease images. It then performs data augmentation on the segmented dataset for training and validation. The MobileNetV2 model, pre-trained on ImageNet is employed for skin cancer classification. The network structure is visualized and specific layers are removed to adapt it to the task. Additional fully connected layers are added to tailor the network for skin cancer classification. The training options including data augmentation and optimization parameters are defined and the model is trained using a specified training set. The code includes loading a pre-trained using a specified training set. It also used for the further analysis and evaluation. Finally it assesses the model performance by calculating accuracy, generating a confusion matrix and plotting a receiver operating characteristic curve. The ROC curve provides insights into the model ability to discriminate between classes. The code is structured to handle skin cancer images dataset and aims to provide an efficient and accurate classification solution leveraging deep learning techniques.