Rapid automated diagnosis of primary hepatic tumour by mass spectrometry and artificial intelligence

Abstract

Background and aims Complete surgical resection with negative margin is one of the pillars in treatment of liver tumours. However, current techniques for intra‐operative assessment of tumour resection margins are time‐consuming and empirical. Mass spectrometry (MS) combined with artificial intelligence (AI) is useful for classifying tissues and provides valuable prognostic information. The aim of this study was to develop a MS‐based system for rapid and objective liver cancer identification and classification. Methods A large dataset derived from 222 patients with hepatocellular carcinoma (HCC, 117 tumours and 105 non‐tumours) and 96 patients with mass‐forming cholangiocarcinoma (MFCCC, 50 tumours and 46 non‐tumours) were analysed by Probe Electrospray Ionization (PESI) MS. AI by means of support vector machine (SVM) and random forest (RF) algorithms was employed. For each classifier, sensitivity, specificity and accuracy were calculated. Results The overall diagnostic accuracy exceeded 94% in both the AI algorithms. For identification of HCC vs non‐tumour tissue, RF was the best, with 98.2% accuracy, 97.4% sensitivity and 99% specificity. For MFCCC vs non‐tumour tissue, both algorithms gave 99.0% accuracy, 98% sensitivity and 100% specificity. Conclusions The herein reported MS‐based system, combined with AI, permits liver cancer identification with high accuracy. Its bench‐top size, minimal sample preparation and short working time are the main advantages. From diagnostics to therapeutics, it has the potential to influence the decision‐making process in real‐time with the ultimate aim of improving cancer patient cure.

New strategy for evaluating pancreatic tissue specimens from endoscopic ultrasound‐guided fine needle aspiration and surgery

“…Thus, we believe that this system has value for diagnosing pancreatic cancer, given the difficulty of collecting a sufficient diagnostic specimen without surgical intervention. This system may also be useful for clinically diagnosing other cancers, including pancreatic, craniofacial, and hepatic cancers 4,5,12 . Finally, it is possible that our system might be useful for predicting outcomes and treatment responses, if the findings can be combined with data regarding pathological findings, TMN staging, chemotherapy outcomes, and prognosis.…”

Lipid Profiling in Cancer Diagnosis with Hand-Held Ambient Mass Spectrometry Probes: Addressing the Late-Stage Performance Concerns

“…Here, averaging replicates and using this value in class differentiations is discouraged as averaging results in loss of valuable information regarding intrasample variability [135]. More interestingly, beyond basic multivariate methods, such as linear discriminant analysis (LDA) [136], least absolute shrinkage and selection operator (LASSO) [137], machine learning approaches such as support vector machine (SVM) [138], and random forest (RF) [139], which require extensive pre-processing of mass spectral data, convolutional neuronal networks have been proposed to offer a higher accuracy of prediction without the need for data pre-processing [140,141]. The predictive power of these methods, however, sharply decreases with small sample sizes [142,143].…”

Proteomic Profiling and Artificial Intelligence for Hepatocellular Carcinoma Translational Medicine

“…The constructed ML-based model showed 74.19% accuracy of the prediction, and the selected mutant genes were verified. Additionally, probe electrospray ionization (PESI) MS in combination with AI have been employed to assess the overall diagnostic accuracy of two algorithms, support vector machine (SVM) and random forest (RF), in HCC detection [ 43 ]. This approach showed bench-top size, minimal sample preparation, and short working time as well as high accuracy, specificity, and sensitivity in HCC diagnosis.…”