Researchers at the University of Missouri are pioneering the use of artificial intelligence to enhance the early detection of melanoma, the most aggressive form of skin cancer. Led by associate research professor Kamlendra Singh, the team is developing decision-support tools that leverage advanced AI models to analyze detailed skin images. This innovative approach aims to identify subtle visual patterns indicative of melanoma with high accuracy, ultimately facilitating quicker diagnosis and treatment, and expanding access to critical dermatological expertise, especially in underserved areas. This research represents a significant step towards integrating AI into precision medicine for patient-centered care.
AI teamwork
In a significant study, Kamlendra Singh and his research team developed and rigorously tested various artificial intelligence models to differentiate between melanoma and benign skin conditions. The training and validation of these AI models utilized an extensive dataset of 400,000 images of skin abnormalities, all obtained through advanced 3D total body photography. This cutting-edge imaging technology provides high-resolution, three-dimensional digital maps of a patient’s entire skin surface, enabling detailed analysis of subtle visual patterns such as size, shape, color, density, and sharpness of moles or suspicious spots. Initially, individual AI models demonstrated an accuracy of up to 88% in identifying melanoma. However, the researchers achieved a substantial improvement in diagnostic precision by combining three distinct AI models, resulting in an impressive accuracy rate exceeding 92%. This ensemble approach leverages the strengths of multiple algorithms to enhance overall performance, moving closer to a reliable AI-powered decision-support tool for dermatologists.
Going forward
Looking ahead, Kamlendra Singh, a principal investigator at the Bond Life Sciences Center, emphasizes the transformative potential of AI in broadening access to critical healthcare services, particularly for populations in regions underserved by specialized medical professionals and equipment. The ongoing refinement of these AI models involves continuous training with increasingly diverse and larger datasets, which will include images reflecting a wide spectrum of skin tones, varied lighting conditions, and different camera angles. This comprehensive approach is expected to significantly enhance the models' predictive accuracy and robustness. While acknowledging that the widespread clinical application of this AI tool is still some time away, Singh regards the current research as a highly promising proof of concept. He stresses the importance for researchers to develop more transparent and explainable AI systems, fostering greater trust among healthcare professionals. Such trust is crucial for the eventual integration of AI as a valuable decision-support tool in clinical practice, ultimately leading to improved patient outcomes through earlier and more accurate diagnoses. Singh attributes the progress of this innovative research to Mizzou’s robust computational infrastructure and the strategic support from the Division of Research, Innovation and Impact, highlighting the university's commitment to translating groundbreaking ideas into tangible solutions. The findings from this pivotal study, titled 'Performance of transformer-convolutional neural network ensemble for melanoma diagnosis on segmented 3D total body photography data: Cross-validation stratified K-fold,' have been published in the respected journal Biosensors and Bioelectronics: X.
