Artificial Intelligence in Hepatology: real promise or hype?

AI is revolutionizing many fields, including hepatology. During EASL 2024, its clinical application was discussed, with a focus on personalized patient management.

AI can handle large amounts of data. But what data?

Artificial intelligence (AI) is an increasingly discussed topic in medicine, and the EASL 2024 congress dedicated a session to its application in hepatology. The central question was whether AI represents real promise or whether, beyond the hype, there is little real substance.

The adoption of AI in healthcare systems faces several difficulties. One of the main obstacles is the lack of adequate electronic health record (EHR) systems in many hospitals. Although patients seem to be ready to use AI, integrating it on a daily basis through fitness gadgets and other technologies, healthcare systems have yet to fully adapt to this innovation.

AI requires high-quality and comprehensive data to work effectively. However, current data sets may be limited and may not include all the data points needed for accurate predictive models. Although the results are promising, the lack of some crucial data points represents a significant challenge.

Artificial Intelligence in hepatology

Prof. Schattenberg identifies 3 areas where artificial intelligence could be used in hepatology:

  1. Data analysis: text mining, clustering analysis, data visualization, machine learning (supervised and unsupervised), deep learning (learning algorithms and interpretation).
  2. Data layer electronic health records, imaging, laboratory, histology data, patient cohorts with clinical outcome in the scientific setting.
  3. Basic layer: cloud computing (availability of platforms and software in the healthcare setting), internet of medical things (access through consumer mobile devices).

AI in clinical practice

A key application of AI is in risk stratification. During the presentation, Prof. Schattenberg reported on various experiences,, with the use of artificial intelligence models to identify people at risk of liver disease at an early stage.1-3

There are high-performance models, which could enable risk stratification, such as the one proposed by Docherty et al for the prediction of NASH. The NASHmap model has achieved high performance, sensitivity and accuracy to detect NASH. However, the clinical application of these models requires the availability of the identified features, which may not always be possible.

AI can improve diagnostic accuracy through imaging analysis. For example, AI algorithms can analyse ultrasound images to detect liver steatosis with high accuracy. However, the quality of the results depends on the standardisation of the images provided to the algorithm.

AI can also improve patient education and personalisation of treatment. For example, an Italian study4 evaluated the answers provided by ChatGPT to patients' questions about liver disease, with generally positive results. This demonstrates the potential of AI as an educational and decision support tool for patients.

AI will transform my clinical practice (and there is work to do)

AI has the potential to transform hepatology by improving diagnosis, management and patient education. However, there are still many challenges to be addressed, including data quality, transparency of algorithms and integration into existing healthcare systems. To fully exploit the benefits of AI, a collaborative approach involving clinicians, researchers and patients is needed.5

More specifically:

  1. Health Care System:
    • harmonize data;
    • adopt data security standard;
    • provide reimbursement;
  2. Research:
    • open source and open data (no black box science);
    • clinical relevance + avoid overinterpretation (endless computational power and huge data sets);
    • quality of data and input;
  3. Physicians:
    • overcome law awareness + expertise;
    • need to standardize data acquisition;
  4. Patients:
    • maximize benefit  - maintain a relation of trust  (man vs machine);
    • education and empowerment.
References
  1. Noureddin M, Ntanios F, Malhotra D, Hoover K, Emir B, McLeod E, Alkhouri N. Predicting NAFLD prevalence in the United States using National Health and Nutrition Examination Survey 2017-2018 transient elastography data and application of machine learning. Hepatol Commun. 2022 Jul;6(7):1537-1548. doi: 10.1002/hep4.1935. Epub 2022 Apr 1. PMID: 35365931; PMCID: PMC9234676.
  2. Schattenberg JM, Balp MM, Reinhart B, Tietz A, Regnier SA, Capkun G, Ye Q, Loeffler J, Pedrosa MC, Docherty M. NASHmap: clinical utility of a machine learning model to identify patients at risk of NASH in real-world settings. Sci Rep. 2023 Apr 5;13(1):5573. doi: 10.1038/s41598-023-32551-2. PMID: 37019931; PMCID: PMC10076319.
  3. Docherty M, Regnier SA, Capkun G, Balp MM, Ye Q, Janssens N, Tietz A, Löffler J, Cai J, Pedrosa MC, Schattenberg JM. Development of a novel machine learning model to predict presence of nonalcoholic steatohepatitis. J Am Med Inform Assoc. 2021 Jun 12;28(6):1235-1241. doi: 10.1093/jamia/ocab003. PMID: 33684933; PMCID: PMC8200272.
  4. Pugliese N, Polverini D, Lombardi R, Pennisi G, Ravaioli F, Armandi A, Buzzetti E, Dalbeni A, Liguori A, Mantovani A, et al. Evaluation of ChatGPT as a Counselling Tool for Italian-Speaking MASLD Patients: Assessment of Accuracy, Completeness and Comprehensibility. Journal of Personalized Medicine. 2024; 14(6):568. https://doi.org/10.3390/jpm14060568
  5. Schattenberg JM. Use of AI in MASLD – promise or hype?. EASL 2024. Thursday, 6 Jun, 08:30 - 09:45 CEST