Early Detection of Superficial Spreading Melanoma With Machine Learning

Overview

Superficial Spreading Melanoma (SSM) accounts for approximately 70% of melanoma cases and can be fatal if not detected early. This project aimed to develop a computer-based method to assist in identifying SSM from dermatoscopic images using machine learning. The hypothesis was that a trained image classification model could accurately distinguish SSM from non-SSM skin lesions, improving early detection and reducing melanoma-related mortality. A binary image classification program was developed using TensorFlow and a convolutional neural network based on the Efficient Net architecture. The model was trained on over 3,000 labeled dermatoscopic images from the RoboFlow “melanoma-new” dataset, supplemented with additional curated images, and classified as SSM or NOT_SSM. Multiple independent control datasets containing thousands of images were reserved exclusively for testing and were not used during training. Model performance was evaluated using accuracy, precision, recall, threshold analysis, and confusion matrices. At a decision threshold of 0.40, the model achieved 81.3% accuracy, 79.2% precision, and 93.65% recall on 6,070 evaluation images. The confusion matrix showed 1,500 true negatives, 902 false positives, 233 false negatives, and 3,435 true positives, demonstrating a clear precision–recall tradeoff. The high recall indicates the model is effective at detecting most SSM cases, which is critical in medical screening where missed cancers are dangerous. Although precision was lower, false positives are less harmful than false negatives because they can be resolved through further medical evaluation. These results support the hypothesis: machine learning can assist in detecting SSM and demonstrate the potential of computer-based pre-screening systems for earlier diagnosis.

Early Detection of Superficial Spreading Melanoma With Machine Learning

Overview

Competition history

Related projects

Detect Early Melanoma Cancer Using Machine Learning

Modeling Deep Learning Neural Net Based Image Recognition to Classify Melanoma Better than a Trained Professional

A Precise, Portable, Non-Invasive Melanoma Detection Device Using Image-Based Deep Learning Approach

A Novel Implementation of Image Processing and Machine Learning for the Early Diagnosis of Melanoma

Utilizing Machine Learning Techniques to Identify Cancerous Skin Lesions

Detection of Melanoma via Deep Learning

Novel Machine Learning Algorithms for Early Diagnosis of Melanoma

Near-Perfect Melanoma Diagnosis with Optimized Feature Learning and Classification

Browse more like this