Data-driven Diagnostics using Multimodal- AI Assisted Approaches for Early Cancer Detection

Project Summary: The vast majority (90%) of cancers are epithelial in nature, and oral squamous cell carcinoma (OSCC) accounts for a portion of these cases by afflicting 744,884 people yearly, worldwide. Late-stage cancer diagnosis is associated with low survival rates, whereas individuals diagnosed at an early stage have a significantly better

chance of survival. Current standard screening examinations often fail to identify abnormal regions with high risk of malignancy, and, thus, need to be improved to increase survival rate. Dr. Gracie Vargas, my sponsor, has developed a detection approach that combines multiple optical imaging modalities to visualize large areas

(widefield, WF) with complementary microscopic areas (nonlinear optical microscopy, NLOM) for label-free identification of neoplasia. This approach has demonstrated substantial image-based alterations in high-risk lesions using the system in preclinical animal models. As with many novel optical systems, our research shows

there is room for optimization in this promising technique. The handling and evaluating of the complex data and identifying the most important features associated with early cancer changes is an additional challenge, particularly in the application in human specimens. Here, I propose an approach that combines label-free

multimodal optical imaging with artificial intelligence (AI) to develop data-driven diagnostics for detection of early high-risk lesions with potential for malignant, to ultimately increase survival rate. In the F99 phase, I will optimize and evaluate a WF system by integrating a seamless spectral capability, to capture a wide range of spectral

features in human OSCC samples. This optimization will acquire additional image-based information that applied to machine learning methods to extract the most important features associated with early cancer changes, while potentially improving the systems performance. My dissertation work during the F99 phase will equip me with

training and expertise in the integration of biomedical optics, under the guidance of Dr. Gracie Vargas at the University of Texas medical Branch at Galveston, and machine learning, with support from my co-sponsor, Dr. Heidi Spratt, to advance translational early cancer detection. In the postdoctoral K00 phase, I will concentrate

on developing multimodal label-free optical approaches for data-driven early cancer diagnostics. This research involves complex decision-making that can extend beyond the capabilities of traditional machine learning. Thus, as I transition to the post-doctoral phase, I will train in advanced AI methods, such as deep learning, as well as

explore the integration of patient records to enhance early cancer diagnostics. The successful completion of this project will advance emerging noninvasive early cancer multimodal microscopy technologies through cutting edge data-driven approaches, ultimately enhancing early epithelial cancer detection and diagnostics.