Hyperspectral Imaging and Machine Learning for Precision Skin Tumor Diagnostics

Our project centers on the innovative integration of machine learning and hyperspectral imaging instrumentation for individual-based skin tumor diagnostics. Our goals are to develop a system that can automatically and accurately delineate skin tumor borders in 3D, with the aim of improving upon current biopsy and excision practices which often require large margins and are prone to incomplete tumor removal. Utilizing advanced machine learning techniques like deep models, active contour algorithms, and image segmentation, we aim to non-invasively predict tumor size from the detailed spectral data captured via different imaging modalities. Our approach combines two imaging modalities, hyperspectral and photoacoustic imaging, to accurately assess tumor spread. Developing our model with real patient data is pivotal in refining our machine learning algorithms to rapidly identify tumor borders in patients with different skin types. The end goal is to create a platform that not only aids surgeons in efficient tumor excision but also facilitates quicker and more accurate diagnoses.