MP17-15: Infrared-(IR)-Imaging Classifies Prostate Cancer Label-free: a Prospective IR-ProSPECT-Study Shows Potential for Clinical Application and Research
Introduction: The Goal of this study was to evaluate whether artificial intelligence (AI) is suited to classify prostate cancer (PCa) label-free. In the last decade, IR-microscopy was able to showcase its potential to classify different tissues and pathologies in several studies. Recorded IR-Spectra reflect the biochemical status of the examined cell. The application of quantum-cascade lasers as source of light has accelerated data acquisition and thus enabled integration into clinical research pathways Methods: In this study, tissue from 258 patients was investigated. Initially, AI with Random Forrest (RF) algorithm was applied. Prostatic tissue showed to be morphologically more demanding than expected. As the study population increased, a robust deep-learning AI was implemented. Results: Deep-learning AI reached an AUC of 0.96 with a sensitivity of ~98% and a specificity of ~83% when classifying PCa label-free. This marks an improvement compared to RF, which showed a sensitivity of 82% and a specificity of 69%. Conclusions: AI-based IR-Imaging detects PCa with a sensitivity of 98%. Further improvement may be achieved by additional training and increase in sample size. Subsequently, laser-assisted micro-dissection may allow precise and homogenous sampling for detailed molecular analysis of tumor tissues. Results may be used to train AI in the detection of molecular tumor characteristics such as oncological targets. SOURCE OF Funding: This project was funded by the Ministry of Culture and Research of the Federal State of Northrine Westphalia
