[F31] High-Resolution Sidestream Dark Field Microscopy for Aiding Diagnosis of Melanoma

Description

PROJECT SUMMARY Melanoma, one of the fastest-growing cancers in the US, has resulted in 8,000 deaths annually, emphasizing the need for early and accurate diagnosis. Standard melanoma diagnosis relies mainly on dermatoscopy-guided biopsies, which have variable diagnostic accuracy and have resulted in up to 96% of all biopsies producing a benign diagnosis. Noninvasive imaging technologies such as Reflectance Confocal Microscopy (RCM) and High- Resolution Optical Coherence Tomography (HR-OCT) provide cellular insight but are limited due to cost and signal specificity. This project addresses those limitations by developing and evaluating a High-Resolution Sidestream Dark Field Microscopy (HR-SDF) to visualize melanocytic cells with high specificity. The proposed study will assess HR-SDF performance in aiding melanoma diagnosis. Through Aim 1, I will compare HR-SDF + dermatoscopy vs. dermatoscopy alone to evaluate the diagnostic performance during melanocytic lesions visualization. Collaborating with experts in the field of novel imaging technology for skin cancer diagnosis, we will conduct a comprehensive imaging study, imaging 200 lesions to develop diagnostic criteria for HR-SDF images. The sensitivity and specificity will be compared to the gold standard pathology report. Through Aim 2, I will explore two deep-learning approaches in parallel for automatically analyzing HR-SDF image stacks: 1) reslice HR-SDF image stacks to produce cross-sectional images, segment these images for melanocytic cells, generate a feature vector from the segmented cells (e.g., depth, color), and feed the feature vector to a machine learning model, for example a support vector machine; 2) feed each expert labeled en face HR-SDF image to a pre-trained feature extraction network to generate the feature map, and feed the feature map to a fully connected network to predict the probability of malignancy. Each approach will undergo five-fold cross-validation to train and evaluate the machine learning model. Beyond the technical aspect, this project will further my career goal of becoming an independent principal investigator, researching and developing innovative imaging technologies. This proposal will train me to conduct a rigorous clinical study of novel imaging technologies, develop advanced deep learning algorithms, and conduct statistical analysis to validate the diagnostic performance of novel imaging technologies. I have built an interdisciplinary team of sponsors and collaborators in the fields of optics, dermatology, and machine learning to ensure this project's success in research and training. Istituzione: UNIVERSITY OF ARIZONA PI: Rafael Romero Progetto: 1F31CA306154-01

Sectors

Source

Source: reporter.nih.gov