Fourier transform InfraRed spectra analyzed by multivariate and machine learning methods in determination spectroscopy marker of prostate cancer in dried serum

Abstract

Prostate cancer represents the second most prevalent form of cancer in males globally. In the diagnosis of prostate cancer, the most commonly utilised biomarker is prostate-specific antigen (PSA). It is unfortunate that approximately 25 % of men with elevated PSA levels do not have cancer, and that approximately 20 % of patients with prostate cancer have normal serum PSA levels. Accordingly, a more sensitive methodology must still be identified. It is imperative that new diagnostic methods should be non-invasive, cost-effective, rapid, and highly sensitive. Fourier transform infrared spectroscopy (FTIR) is a technique that fulfils all of the aforementioned criteria. Consequently, the present study used FTIR to assess dried serum samples obtained from a cohort of prostate cancer patients (n = 53) and a control group of healthy individuals (n = 40). Furthermore, this study proposes FTIR markers of prostate cancer obtained from serum. For this purpose, FTIR spectra of dried serum were measured and analysed using statistical, chemometric and machine learning (ML) algorithms including decision trees C5.0, Random Forest (RF), k-Nearest Neighbours (kNN) and Support Vector Machine (SVM). The FTIR spectra of serum collected from patients suffering from prostate cancer exhibited a reduced absorbance values of peaks derived from phospholipids, amides, and lipids. However, these differences were not statistically significant. Furthermore, principal component analysis (PCA) demonstrated that it is challenging to distinguish serum samples from healthy and non-healthy patients. The ML algorithms demonstrated that FTIR was capable of differentiating serum collected from both analysed groups of patients with high accuracy (values between 0.74 and 0.93 for the range from 800 cm⁻¹ to 1800 cm⁻¹ and around 0.70 and 1 for the range from 2800 cm⁻¹ to 3000 cm⁻¹), depending on the ML algorithms used. The results demonstrated that the peaks at 1637 cm⁻¹ and 2851 cm⁻¹ could serve as a FTIR marker for prostate cancer in serum samples. Furthermore, the correlation test indicated a clear correlation between these two wavenumbers and four of the five clinical parameters associated with prostate cancer. However, the relatively small number of samples collected only from patients over the age of 60 indicated that the results should be further investigated using a larger number of serum samples collected from a mean age range. In conclusion, this study demonstrated the potential of FTIR for the detection of prostate cancer in serum samples, highlighting the presence of distinctive spectroscopic markers associated with the analysed cancer type.