Given that routine therapies are rarely curative in the occurrence of distant metastasis, cancer, which can originate from different site of the body, such as lung, stomach, bladder, skin etc., is considered one of the most fatal diseases. As one of the top 6 most frequently diagnosed cancers that lead to fatal death [1], esophageal cancer has two major histological sub-types, i.e., esophageal adenocarcinoma (EAC) and esophageal squamous cell carcinoma (ESCC). EAC mainly occurs in the western countries, whereas ESCC is more prevalent worldwide with a high incidence in those developing countries in Asia and south-eastern Africa. Despite the epidemiological differences, both types of oesophageal cancer have a poor prognosis, and profound improvements can be seen if they are treated prior to invasion [2,3]. In this case, one major goal in esophageal cancer research is to develop a rapid, cheap and sensitive detection method for the diagnosis before the cancer metastasizes to distant sites.

Among all the current detecting approaches for the diagnosis of oesophageal cancer and preinvasive lesions, endoscopy is considered the gold-standard technique owning to its high sensitivity [4]. However, considering the high cost, the invasiveness of the procedure and the psychological burden, it is not accepted as an excellent national screening program [5]. Tissue-based tumor profiles are, generally speaking, subject to sampling bias, lack of repeatability, inconvenience of operation and invasiveness to human body [6]. For example, the esophageal balloon cytology method has been extensively used for detecting early manifestations of cancer in patients. The current blind sampling method of a large organ may miss small lesions and rare abnormal cells. Moreover, it can easily cause heartburn symptoms for patients. Therefore, current precision oncology has increasingly paid attention to liquid biopsies, since they are non-invasive and can be repeated at multiple time points [7].

Plenty of substances that have been found in the peripheral blood make blood the most used fluids in liquid biopsies. These substances are closely related to tumor stage and might serve as new biomarkers, including circulating tumor cells (CTCs), exosomes, circulating tumor DNAs (ctDNAs), and a series of cancer-related proteins. However, the complexity of blood samples and the extremely low concentration of the analytes challenge the sensitivity and reliability of current detection methods for clinical applications [8]. Fourier-transform infrared (FTIR) spectroscopy is a well-established optical technique capable of providing the biochemical information of the samples on the molecular level. It has multiple advantages, including high sensitivity, low cost, high repeatability, rapid and non-destructive, which renders it a perfect candidate for clinical translation. Up to now, using blood serum and plasma, FTIR has been employed to diagnose rheumatoid arthritis [9], HIV/AIDS [10], leukemia [11], depression [12], alzheimer's [13], malaria parasitemia [14,15], glucose [16], breast cancer [17], [18], [19], [20], gastric cancer [21], liver cancer [22], ovarian cancer [23], lung cancer [24], endometrial cancer [25], bladder cancer [26,27], brain tumor [28,29], oral submucous fibrosis, colitis screening [30,31], etc.

As a potential clinical detecting method, FTIR technique has been used for discriminating esophageal cancer cell [32], [33], [34], [35] and Barrett's esophagus tissues [36], [37], [38], [39]. For biofluids detection, a recent study has applied the attenuated total reflection (ATR) FTIR to the interrogation of the oesophageal stages of transformation to EAC, employing the biofluids plasma, serum, saliva, or urine [40]. However, the transition from FTIR research to routine clinical testing remains elusive and is exposed to several major technological challenges, such as the disability to identify the disease type, lack of standardization of the technique, and validation of large cohort of samples etc. [41,42]. In addition, it is widely acknowledged that the current esophageal cancer study using FTIR mainly focus on EAC, with few on ESCC, making it of great significance of value to establish a rapid, cheap and sensitive detecting system employing FTIR technique in ESCC diagnosis for developing countries.

In this study, exploration was conducted on the detection of esophageal squamous cell carcinoma (ESCC) using serum and plasma by transmission FTIR method. Customized Barium fluoride (BaF2) slices enable high-throughput FTIR measurement for clinical research. The changes which occur between FTIR spectra of different stage cancer samples can often be small and not easily artificial observable, therefore machine learning algorithms are frequently used to differentiate these differences in spectral data [43], [44], [45], [46], [47]. Decision Tree (DT), Support Vector Machine (SVM), k-Nearest Neighbour (KNN) classification, Ensemble Algorithms (EA), Partial Least Squares for Discriminant Analysis (PLS-DA) and Principal Component Linear Discriminant Analysis (PC-LDA) were used to discriminate the FTIR spectral profiles in this paper. Additionally, previous study has shown that the choice of anticoagulant is noticeable in the resultant infrared spectrum [14], which may influence the diagnostic performance. To this end, the diagnostic performance by serum and plasma with different anticoagulants was also assessed.