AI Tool Achieves 99% Accuracy in Melanoma Detection
Northeastern University’s AI tool for melanoma detection achieves exceptional accuracy.
- • Northeastern University developed SegFusion Framework for melanoma detection
- • Tool achieves 99.01% accuracy using deep learning technologies
- • Plans to integrate patient records and develop a dermatologists' app
- • Potential future applications for other types of cancer detection
Key details
Researchers at Northeastern University have announced the development of a pioneering AI tool capable of detecting melanoma with 99% accuracy, addressing the pressing need for timely identification of this serious skin cancer. As the American Academy of Dermatology predicts approximately 212,000 new melanoma cases in the U.S. by the end of 2025, early detection becomes critical to improving patient outcomes and lives.
The innovative system, known as the SegFusion Framework, was engineered by Divya Chaudhary and graduate student Peng Zhang. This hybrid model uniquely combines the strengths of U-Net and EfficientNet deep learning architectures, allowing for a two-step analysis of skin images: it first identifies suspicious areas and then evaluates these to ascertain if they indicate cancerous lesions. During tests on the International Skin Imaging Collaboration 2020 dataset, the SegFusion Framework achieved a remarkable accuracy of 99.01%, outshining previous models such as ResNet-101+SVM and MobileNetV2, which had accuracies of 97.15% and 98.2%, respectively.
Chaudhary pointed out the potential for the framework to significantly enhance not just melanoma diagnoses but also broader cancer detection efforts. The researchers trained the model using two key dermatology image datasets: HAM10000 and ISIC 2020. To improve learning efficacy, they strategically oversampled melanoma cases, which constituted only 1.8% of the ISIC dataset. A novel 'data bridge' connects the segmentation and classification models, facilitating a cohesive analysis of potentially cancerous regions.
Future plans for the SegFusion Framework include the integration of patient health records to further refine its accuracy and the development of an app designed for real-time consultation by dermatologists during examinations. Chaudhary envisions the possibility of adapting this AI technology for diagnosing other cancers, such as breast and lung cancer, thereby expanding its impact in the field of medical diagnostics.