A pilot study to implement an artificial intelligence (AI) system for gastrointestinal cancer clinical trial matching

Abstract

Background: Hepatocellular carcinoma (HCC) has relatively sensitive and specific serum tumor antigen markers (AFP), which is also the most common serological marker for cancer screening. However, there are unignorable limitations, including possible false-negatives/positives owing to confounding conditions. Reliable non-inva-sive diagnostics is still in urgent need. This work proposes a novel LDI-TOF-MS technique for HCC screening and diagnosis. By taking advantage of 3D nanostructures and machine learning, our technique enables high fidelity and reproducibility. Methods: An LDI-TOF-MS platform was established for HCC screening and was applied to 139 patients with liver cancer, as well as 203 healthy controls (Table). All mass spectrum was collected within a mass range of 100 to 1,100 Da for metabolites. Based on the data acquired by LDI-TOF-MS, SVM algorithm was developed and applied for automated cancer classification across six cancer types, which was further validated by single blinded samples with randomly selected cancer patients and controls.