Attenuated total reflection-Fourier transform infrared spectroscopy (ATR-FTIR) of liquid biofluids enables the probing of biomolecular markers for disease diagnosis, characterized as a time and cost-effective approach. It remains poorly understood for fast and deep diagnosis of digestive tract cancers (DTC) to detect abundant changes and select specific markers in a broad spectrum of molecular species. Here, we present a diagnostic protocol of DTC in which the in-situ blood-based ATR-FTIR spectroscopic data mining pathway was designed for the identification of DTC triages in 252 blood serum samples, divided into the following groups: liver cancer (LC), gastric cancer (GC), colorectal cancer (CC), and their different three stages respectively. The infrared molecular fingerprints (IMFs) of DTC were measured and used to build a 2-dimensional second derivative spectrum (2D-SD-IR) feature dataset for classification, including absorbance and wavenumber shifts of FTIR vibration peaks. By comparison, the Partial Least-Squares Discriminant Analysis (PLS-DA) and backpropagation (BP) neural networks are suitable to differentiate DTCs and pathological stages with a high sensitivity and specificity of 100% and averaged more than 95%. Furthermore, the measured IMF data was mutually validated via clinical blood biochemistry testing, which indicated that the proposed 2D-SD-IR-based machine learning protocol greatly improved DTC classification performance.

In machine learning, traditional single datasets often cause confusion between species, mixing features to form a mixed dataset will solve this challenge.

No abstract

No abstract

No abstract