Abstract
Histological staining has long been the gold standard for cancer detection, but it is limited by subjectivity and delayed results. Fourier Transform Infrared Spectroscopy (FTIR) has emerged as a promising technique, offering the advantages of objectivity and real-time analysis. Despite its potential, IR biomarkers developed in different studies are based on varying experimental conditions, including different tissue preparation methods, instrumentation, and patient heterogeneity, which hinder their generalizability and transferability. In this study, tissue spectra from various colorectal cancer cases were systematically collected, and IR biomarkers were developed using the band ratios identified in these spectra. Their ability to differentiate cancer-bearing from non-cancer-bearing tissues was evaluated using a machine learning technique. The results show that the top three biomarkers are b1 (1740/1236), b2 (1740/1162) and b3 (1740/1080). When transferring these biomarkers to a new case (approved under the Ohio State University Institutional Review Board # 2011C0085), b1 effectively differentiates tumor, normal, and margin tissues, while the performance of the b2 and b3 are less satisfactory. This suggests that b1 is more robust and sensitive to key molecular features, whereas b2 and b3 are more likely to be affected by tissue heterogeneity or experimental variations. Combining biomarkers enhances tissue differentiation, but the enhancement plateaus after two to three key biomarkers, as adding more introduces overlapping or redundant information. This study provides a benchmark for future research, with the aim of advancing the clinical translation of noninvasive infrared spectroscopy for intraoperative applications.