Identification of colorectal malignancies enabled by phasor-based autofluorescence lifetime macroimaging and ensemble learning
Joao L. Lagarto,Alberto I Herrando,Rafaela Rego,Laura Fernandez,Jose Azevedo,Hugo Domingos,Pedro Vieira,Amjad Parvaiz,Vladislav I. Shcheslavskiy,Pedro G. Silva,Mireia Castillo-Martin
DOI: https://doi.org/10.1101/2024.12.04.24317691
2024-12-05
Abstract:Colorectal cancer (CRC) remains one of the most frequent cancers and a leading contributor to cancer-associated mortality globally. CRCs are often diagnosed at an advanced stage, which leads to high mortality and morbidity. This outcome is exacerbated by high rates of recurrence and postoperative complications that contribute substantially to poor prognosis. Advancements in endoscopic assessment have improved CRC prevention, early detection, and surveillance over the years. Yet, CRC remains one of the most significant health challenges of the 21st century. Label-free optical spectroscopy methods have long been explored as potential partners to endoscopy, not only to enhance diagnostic accuracy but also to confer predictive capabilities to endoscopic evaluations. In this study, we investigated the potential of time-resolved autofluorescence measurements excited at 375 nm and 445 nm to correctly classify benign and malignant tissues in CRC surgical specimens from 117 patients. Multiparametric autofluorescence lifetime data were collected in two distinct datasets, which were used for training (n = 73) and testing (n = 44) a supervised classification model, with standard histopathology assessment serving as ground truth. Using 5-fold cross-validation, we achieved 82.6 +- 0.02% sensitivity, 90.4 +- 0.01% specificity, 87.4 +- 0.01% accuracy, and 0.941 +- 0.004 area under the curve (AUC) for training data. Evaluation on unseen test data yielded similar results, with 85.2% sensitivity, 84.5% specificity, 84.8% accuracy, and 0.915 AUC. While preliminary, our findings underscore the potential impact of AI-assisted autofluorescence lifetime measurements in advancing CRC prevention, early detection, and surveillance efforts.