부산대학교 도서관

Introduction:Fibrosis is significantly associated with nearly all forms of heart and kidney disease. Clinical diagnosis of fibrosis is currently reliant on conventional methods that do not have the sensitivity and specificity required for effective diagnosis.Hypothesis:An handheld portable near-infrared (NIR) spectrometer, usable intraoperatively coupled to machine learning algorithms can discriminate between fibrotic and healthy cardiac and renal tissue. We sought to validate this in an animal model.Methods:10 Male Sprague Dawley rats (SDR) with either induced cardiac (SDR-H) and renal (SDR-K) fibrosis (n=5) compare to normal controls (n=5). Hearts from all rats were used as tissue screening in model validation as they contain a high amount of collagen. Multiple tissue samples were harvested from SDR-K (nfibrosis= 12, ncontrol= 4) and SDR-H (nfibrosis= 12, ncontrol= 4) groups. NIR spectra (1350 - 2500 nm) were acquired from all tissue sections.Results:Stained sections showed insignificant differences between the fibrotic SDR-H and their corresponding controls as collagen fibrils dominated both groups. SDR-K showed distinguishable features between examined groups. NIR absorption at 1509, 1725, 2055, and 2306 nm were found to be highly indicative of fibrosis (Figure1). PCA (57% explained variance) and PLS-DA (sensitivity: 96%) showed excellent discrimination for SDR-K groups while the heart shows no meaningful discrimination for SDR-H groups. SVM and LR analysis corroborated these results by achieving a 98% classification accuracy for SDR-K and no discrimination for SDR-H. All machine learning models were cross-validated with outcomes of histological staining to establish a robust interpretation and underpin their pathological meanings.Conclusions:NIR accurately diagnoses cardiac and renal fibrosis in rats model. There is potential for this technology to be translated into an intraoperative instrument for tissue diagnosis