Prediction of Recurrence Following HCC Resection Using Artificial Intelligence: A systematic review of 23,693 patients

Wednesday, March 8, 2023

4:12 PM – 4:21 PM

Location: Phoenix Convention Center, 222AB

Presenting Author(s)

Pedram Keshavarz, MD

Postdoc Research Fellow
David Geffen School of Medicine At UCLA

Disclosure(s): No financial relationships to disclose

Author/Co-author(s)

FY

Fereshteh Yazdanpanah, MD

Postdoc Research Fellow
Department of Diagnostic & Interventional Radiology of New Hospitals LTD, Tbilisi, Georgia
AH

Andrew D. Holmes, BS, BA

Medical Student
University of South Dakota Sanford School of Medicine
NC

Nicholas G. Cen, n/a

Medical Student
David Geffen School of Medicine At UCLA
MB

Mansoureh Baradaran, MD

Radiologist
Dep. of Diagnostic & Interventional Radiology at New Hospitals LTD
KS

Kyung H. Sung, PhD

Associate Professor of Radiology
David Geffen School of Medicine At UCLA
David S. Lu, MD (he/him/his)

Professor of Radiology, and Surgery
David Geffen School of Medicine at UCLA
Steven Raman, MD, FSAR, FSIR

Professor of Radiology, Urology and Surgery
David Geffen School of Medicine At UCLA

Disclosure(s):

Pedram Keshavarz, MD: No financial relationships to disclose

David S. Lu, MD: Ethicon J&J: Advisory Committee or Review Panel Member (Ongoing), Consultant (Ongoing), Speaking and Teaching (Ongoing)

Purpose: The main purpose of this study is to review artificial intelligence algorithms such as machine learning (ML) and deep learning (DL) that use radiomic features to predict the early recurrence of hepatocellular carcinoma (HCC) after resection.

Materials and Methods: A systematic search was performed in Scopus, Pubmed, Embase, Google Scholar, and Web of Science. The reported accuracy, predictive values, and AUC from individual studies were derived, and the results were evaluated based on the “Standards for Reporting Diagnostic accuracy studies” (STARD) guidelines. Study subcohort categories were as follows: ML-radiomics, DL-radiomics, ML-non-radiomics, and DL-non-radiomics model.

Results: Overall, 39 studies (23,693 overall patients), including 22 (1,492 patients, validation group), 3 (204 patients, validation group), 10 (5,849 patients, validation group), 4 (420 patients, validation group) in ML-radiomics, DL-radiomics, ML-non-radiomics, and DL-non-radiomics predicted the recurrence following HCC resection, respectively. The overall accuracy was 0.71 (95%CI, 0.69-0.76), 0.75 (95%CI, 0.71-0.78), 0.64 (95% CI, 0.61-0.67) and 0.71 (95%CI, 0.54-0.88), respectively and AUC was 0.71 (95%CI, 0.69-0.76), 0.77 (95%CI, 0.71-0.81), 0.59 (95%CI, 0.49-0.69), and 0.72 (95%CI, 0.63-0.74) for ML-radiomics, DL-radiomics, ML-non-radiomics, and DL-non-radiomics models, respectively. C-indexes of models, which measure accuracy of recurrence prediction were 0.7 (95% CI, 0.64-0.73) and 0.66 (95% CI, 0.64-0.67) in the ML-radiomics model (1,358 patients) and ML-non-radiomics model (12,475 patients). Time to recurrence (TTR) was reported to be 18 months (range, 7.5-38.2) in the four studies (272 patients).

Conclusion: The combined models integrating ML or DL with radiomics features extracted from imaging modalities may be a potential imaging biomarker for non-invasively identifying patients at significant risk of early HCC recurrence following resection.