Identification of Hepatic Fibrosis and Steatosis via A Point-of-Care Transient Elastography System With Integrated AI

Transient elastography (TE) is routinely undertaken for non-invasive assessment of liver fibrosis and steatosis, but is limited by its bulky design, inadequate imaging guidance and conventional algorithmic framework. Thus, we report a real-time B-mode image-guided, artificial intelligence-assisted, point-of-care TE (AI-POC-TE) system, providing simultaneous liver stiffness measurement (LSM) and a novel multi-domain attenuation parameter (MAP) for fat quantification. We aimed to determine the accuracy of LSM and MAP in diagnosing histology-confirmed fibrosis and steatosis in patients with chronic liver disease. Exploratory analyses assessed the minimum number of measurements required.

This prospective study included 138 patients who underwent liver biopsy and AI-POC-TE simultaneously, and diagnostic performance was evaluated by area under the receiver operating characteristic curve (AUROC). Another larger cohort of 1455 patients was examined to benchmark AI-POC-TE against conventional TE (Fibroscan).

LSM by AI-POC-TE identified patients with fibrosis with AUROCs of 0.79 for ≥F2, 0.79 for ≥F3, 0.97 for F4. Corresponding Youden's cut-offs were 8.2, 9.1 and 14.4 kPa. MAP detected steatosis of ≥ S1, ≥ S2, S3 with AUROCs of 0.92, 0.70, 0.76 and Youden's cut-offs were 244, 278 and 294 dB/m, respectively. Among 1455 patients using both TE techniques, liver stiffness was highly correlated (r = 0.86) and MAP also correlated well with CAP (r = 0.80). Fewer than 10 measurements suffice to maintain accuracy; four measurements were statistically non-inferior to the standard 10, supporting a streamlined protocol.

We found AI-POC-TE to accurately assess fibrosis and steatosis, comparable to conventional TE but with added values of portability, B-mode guidance and deep learning-based analytics.