Jim Sweeney discusses a novel test that can help pathologists diagnose MASH
Metabolic dysfunction-associated steatohepatitis (MASH) is an advanced stage of metabolic dysfunction-associated steatotic liver disease (MASLD) – and these two conditions are fast becoming healthcare priorities. MASLD is estimated to impact up to 37 percent of the global population, with 5–6 percent of the US population observed to progress to MASH (1, 2, 3).
Jim Sweeney, President of PathAI Diagnostics, tells us about AI.Dx MASH – a first-of-its-kind test that provides AI-assisted histologic scoring to help pathologists improve outcomes for patients with MASH.
Could you describe the potential benefits of early and accurate diagnosis of MASLD and MASH?
Currently there are no approved therapeutics for the treatment of MASH. As a result, lifestyle changes aimed at reducing weight (for example, diet and exercise) are the main mechanisms to address the condition. The earlier a patient can be informed of their diagnosis and act upon it, the less likely they are to progress to cirrhosis, where changes in liver biology are significantly more difficult to reverse. As therapeutics are approved, there could be opportunities to associate liver biopsy scoring performed via this sort of test with the resulting therapeutic outcomes, but this would require further studies. Ultimately, the economic impact of such testing would be in helping to halt disease progression through timely and accurate intervention, potentially facilitating a lesser need for transplantation or cancer treatment resulting from disease progression to cirrhosis.
Could you provide more information about the AI.Dx MASH test?
AI.Dx MASH is a laboratory developed test (LDT) that uses liver biopsy samples that are sent to our pathology laboratory in Memphis, Tennessee. When MASH is suspected, these biopsies are processed and undergo H&E and Masson’s Trichrome staining before the slides are digitized on-site. The digital image can then be processed using the AI-powered algorithm that provides the basis for AI.Dx MASH. The algorithm was originally created using a library of over 5,900 liver biopsies containing over 100,000 annotations across a range of MASH cases. These cases were reviewed and annotated by a cross section of the company’s pathologist contributor network, meaning that the resulting learning data set truly comprised inputs from a range of fellowship-trained pathologists.
Leveraging these inputs – which include both annotations of histologic features and disease severity scores – the algorithm learned to not only detect and quantify a range of histologic features from liver biopsy tissue, but also to perform histologic scoring in alignment with the NASH Clinical Research Network (CRN)’s guidelines. AI.Dx MASH provides the AI-derived NASH CRN ordinal grades for steatosis, lobular inflammation, and hepatocellular ballooning, in addition to stages for fibrosis, which can then be reviewed by our fellowship-trained gastrointestinal and hepatobiliary pathologists at PathAI Diagnostics prior to delivery of a final report.
In addition, the algorithm applies a colorized overlay on top of the tissue image. Both the whole slide image and overlay are visualized via PathAI’s proprietary image management system, AISight, and the overlay spotlights histologic features that are relevant to MASH grading and staging decision-making. AI.Dx MASH also provides the reviewing pathologist with quantitative data for relevant histologic components with respect to the percentage of the sample containing steatosis, and so on. Such quantifications have been shown in scientific literature to be highly relevant to understanding MASH disease severity, in addition to progression and regression; however, their clinical validity is still to be determined. Importantly, these quantitative features provide more detail than healthcare providers are used to seeing. Ultimately, in reviewing a case using AI.Dx MASH, the pathologist can confirm if they agree with the algorithm’s scoring before generating a report that goes to the requesting physician.
Why did PathAI Diagnostics decide to develop the world's first AI-assisted laboratory test for MASLD and MASH?
Variability in liver biopsy results is a key challenge in the diagnosis and scoring of MASLD/MASH. There can be more than 30 percent discordance between pathologists on whether a patient does or does not have MASH, and inter- and intra-observer scoring variability has been shown to negatively impact MASH clinical trial outcomes (4). The literature shows that inter-reader agreement in any one of these measurements can be as low as 31 percent, and even the same reader looking at the same case may change their score up to 45 percent of the time (4).
The NASH CRN scoring system uses thresholds that can prove challenging, resulting in variability near the boundaries. For example, with steatosis, a score of one or two can occur because of a one percent difference (33 percent versus 34 percent steatosis). Equally, hepatocellular ballooning is scored based on a measurement of “few” versus “many” cells, and this subjectivity can also contribute to scoring variability. The end result is that MASH histologic scoring can be quite “noisy” data.
The potential impact of this “noise” has been demonstrated in clinical studies run for MASH Scoring variability and makes patient cohort identification and therapeutic response measurement challenging. The algorithm used in AI.Dx has demonstrated the ability to resolve differences in a clinical trial setting, so our hope is that this increased consistency in scoring can also be leveraged to provide benefit for patients in general.
How will the introduction of AI-assisted histologic scoring impact the accuracy of liver biopsy reporting and patient outcomes for individuals with MASLD and MASH?
We have been very deliberate in highlighting the AI-assistance in this LDT because we do not want users to think of this as a replacement for traditional pathology or pathologists – it is something that is intended to augment the quality of the histologic report resulting from a biopsy. Indeed, the introduction of AI.Dx MASH provides an additional objective observer for MASH/MASLD liver biopsies. The algorithm has been used for several clinical studies for MASH where it was challenging to interpret observed differences in drug responses between placebo and treatment groups because of the aforementioned variability in histologic scoring. Because the algorithm was trained leveraging a large number of cases in addition to annotations provided by a group of experts, it captures a general consensus across liver pathologists about how these histologic features should be evaluated and scored, rather than reflecting an individual pathologist’s scoring habits. As such, we hope that this tool can offer guidance, especially to GI pathologists who may not have as much experience evaluating MASH biopsies specifically. PathAI continues to publish data in this space and this helps to reinforce the value of AI in helping to address where a patient currently resides on the continuum of the disease.
The significance of AI.Dx MASH in a clinical setting lies in its ability to influence patient outcomes through the recognized correlation between the histologic score assigned to MASH and downstream effects. One study showed that, as a patient progresses from stages I to stage V of fibrosis, the risk of liver-related mortality doubles, and there are similar increases in risk associated with liver transplantation (5). Understanding more precisely where a patient is on the disease spectrum through histologic scoring provides an opportunity to address the condition.
What are the key milestones for PathAI Diagnostics in the deployment of AI.Dx MASH and development of other AI-assisted diagnostic tools?
We have already reached the first critical milestone in validating AI.Dx MASH as an LDT, and we are eager to have the data this test produces in the hands of physicians, so they are positioned to provide the best care for their patients. Beyond AI.Dx MASH, we are looking to develop a portfolio of AI-assisted tests that will all work to improve patient outcomes by providing the referring physician with less variability and more robust diagnostic data.
Given the rapidly evolving field of medical AI, how does PathAI Diagnostics ensure the safety, privacy, and ethical use of patient data?
We are no strangers to the importance of respecting the safety and privacy of patient data. We uphold HIPAA in the handling of patient data throughout the receipt, processing, and reporting of results. In terms of safety, all of the outputs generated by the AI in AI.Dx MASH can be reviewed and over-ridden by one of our board-certified pathologists, so we do not allow the AI to make unchecked determinations. In terms of ethical consideration, we see the use of AI.Dx MASH as a critical component in improving the accuracy of information provided to the healthcare provider and indirectly the patient. The use of AI.Dx MASH is an important tool to improve accuracy, support clinical decision making – and ultimately – better serve patients.
- K Riazi et al., Lancet Gastroenterol Hepatol, 7, 851 (2022). PMID: 35798021
- C Estes et al., Hepatology, 67, 123 (2018). PMID: 28802062
- A M Diehl et al., N Engl J Med, 377, 2063 (2021). PMID: 29166236
- B Davison et al., J Hepatol, 73, 1322 (2020). PMID: 32610115
- RS Taylor et al., Gastroenterology, 158, 1611 (2020). PMID: 32027911
Associate Editor for the Pathologist