A recent study published in the journal Nature Medicine has introduced an innovative artificial intelligence (AI) digital pathology tool called "AI-based Measurement (AIM) for Metabolic Dysfunction-Associated Steatohepatitis (AIM-MASH)".
AIM-MASH H&E and Trichrome inference pipelines. H&E, hematoxylin and eosin; TC, trichrome. Image Credit: https://www.nature.com/articles/s41591-024-03172-7
This system is designed to assess the histological features of MASH, previously known as non-alcoholic steatohepatitis (NASH) accurately and consistently. The researchers aim to enhance the reliability of liver evaluations in MASH clinical trials, which is essential for developing effective treatments.
Background
MASH is a liver disease characterized by inflammation and scarring of liver tissue. It is a leading cause of liver-related illness and death worldwide, often linked to obesity, diabetes, and other metabolic conditions. Diagnosing MASH typically requires a liver biopsy, which a pathologist reviews to determine the extent of liver damage. However, this process can be subjective and inconsistent, leading to variability in diagnosis and treatment plans. In clinical trials, liver biopsies must be histologically scored to meet inclusion criteria and assess endpoints.
Recently, AI has shown promise in enhancing the accuracy and efficiency of medical diagnoses. AI systems can process large datasets, including medical images, to identify patterns and make predictions. In liver disease, AI has been used to develop systems that analyze liver biopsies and provide accurate diagnoses.
About the Research
In this paper, the authors developed "AIM-MASH" to assess liver histology in MASH patients. Their algorithm uses multiple convolutional neural networks (CNNs) and graph neural networks (GNNs) to generate various histological readouts.
The CNN models were trained and validated on a large dataset of liver biopsies from MASH patients, annotated by expert pathologists. This dataset included over 103,579 pathologist annotations from 8,747 hematoxylin and eosin (H&E) and 7,660 Masson's trichrome (MT) whole-slide images (WSIs) from six completed phase 2b and phase 3 MASH trials. The models were designed to segment and quantify key histological features, such as macrovesicular steatosis, lobular inflammation, hepatocellular ballooning, and fibrosis.
The GNN models were then trained to predict MASH clinical research network (CRN) ordinal grades or stages and corresponding continuous scores for each core histological feature of MASH. This two-stage machine learning approach generated patient-level predictions of MASH CRN component scores and fibrosis stage.
Research Findings
The outcomes showed that the AIM-MASH algorithm provided perfectly repeatable scores, achieving a 100 % model-to-model agreement rate (κ = 1) for the four key histological features. In contrast, previous methods showed variable intra-pathologist agreement, ranging from 37 % to 74 %.
When comparing AIM-MASH predictions to a pathologist-based consensus, the agreement rates were consistent with those reported for inter-pathologist comparisons. The highest agreement with the consensus was for steatosis (κ = 0.74), followed by ballooning (κ = 0.70), lobular inflammation (κ = 0.67), and fibrosis (κ = 0.62). Importantly, the agreement between AIM-MASH and the consensus exceeded the agreement observed between any individual pathologist and the other three pathologists, as well as the average pairwise pathologist agreement.
The researchers further validated the tool's clinical utility by evaluating patient eligibility for MASH clinical trials and assessing key endpoints. AIM-MASH showed comparable or even superior agreement rates compared to pathologists when distinguishing between various stages of MASH and fibrosis. In a retrospective analysis of the ATLAS trial, which examined the effects of adjuvant tamoxifen, AIM-MASH identified a higher placebo-adjusted response rate for crucial histological endpoints than the central reader. This suggests that AIM-MASH may be more sensitive in detecting treatment effects.
Applications
The AIM-MASH system has several potential implications for liver disease diagnosis and treatment. It can enhance the accuracy and efficiency of liver biopsy evaluations, leading to more consistent diagnosis and treatment plans. The system can also track changes in liver histology over time, helping to monitor treatment effectiveness. Additionally, AIM-MASH can identify patients at high risk of developing liver disease and monitor their progress. This can help to prevent liver disease and improve patient outcomes.
Conclusion
In summary, the novel AIM-MASH system effectively improved the accuracy and efficiency of liver disease diagnosis and treatment. It could potentially revolutionize digital pathology by offering a reliable, quantitative, and reproducible assessment of MASH histology. This AI-powered tool addressed the variability of manual histological evaluations, leading to more consistent measurements of disease severity and treatment response. Integrating this system into clinical trial workflows could accelerate the development of life-changing treatments for patients with this challenging liver disease.
Journal Reference
Iyer, J.S., Juyal, D., Le, Q. et al. AI-based automation of enrollment criteria and endpoint assessment in clinical trials in liver diseases. Nat Med (2024). DOI: 10.1038/s41591-024-03172-7, https://www.nature.com/articles/s41591-024-03172-7
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.