Real-Time, On-Site, Machine Learning Identification Methodology of Intrinsic Human Cancers Based on Infra-Red Spectral Analysis – Clinical Results
In this work we present a real-time (RT), on-site, machine-learning based methodology for identifying intrinsic human cancers. The presented approach is reliable, effective, cost-effective and non-invasive and based on the Fourier transform infrared (FTIR) spectroscopy - a vibrational method with the ability to detect changes as a result of molecular vibration bonds using infrared (IR) radiation in human tissues and cells.
Medical IR optical system (IROS) is a table-top device for real-time tissue diagnosis that utilizes FTIR spectroscopy and the attenuated total reflectance (ATR) principle to accurately diagnose the tissue. The ATR measurement principle is performed utilizing a radiation source and a Fourier transform (FT) spectrometer. Information acquired and analyzed in accordance with this method provides accurate details of biochemical composition and pathologic condition of the tissue.
The combined device and method were used for RT diagnosis and characterization of normal and pathological tissues ex-vivo/ in-vitro. Therefore, the presented device can be used in close conjunction with a surgical procedure
The solution methodology is to select a set of "features" that can be used to differentiate between cancer, normal and other pathologies using an appropriate classifier. These features serve as spectral signatures (intensity levels) at specific values of measured FTIR-ATR spectral responses.
Excellent results were achieved by applying the following three machine learning (ML) based classification methods to 76 wet samples: Partial least square regression (PLSR) and Principal component regression (PCR)
Both of the methods (PCR & PLSR) show a high performance to classify "Cancer" or "non-Cancer"; Correct Classification: 100 %; Incorrect Classification: 0.0 %.
Naive Bayesian classifier (NBC); Shows a high performance to classify "Cancer" or "non-Cancer" (benign); Correct Classification: 100 %; Incorrect Classification: 0.0 %.