AI-powered liver disease diagnosis
Machine learning for treatment planning
Predicting disease progression
The future of hepatology
References
Further reading
Artificial Intelligence (AI) is an umbrella term that covers all computational processes aimed at mimicking and extending human intelligence for problem-solving and decision-making. It is based on algorithms or arrays of mathematical formulae that make up specific computational learning methods. Machine learning (ML) and deep learning (DL) use algorithms in more complex ways to predict learned and new outcomes.
Liver Disease Diagnosis and Treatment" />
Image Credit: Marko Aliaksandr/Shutterstock.com
Hepatology largely depends on imaging, a field that AI can fully exploit. Machine learning is being pressed into play to extract rich information from imaging and clinical data to aid the non-invasive and accurate diagnosis of multiple liver conditions.
AI can also help develop objective risk stratification scores, predict the course of disease or treatment outcomes in CLD or liver cancer, facilitate easier and more successful liver transplantation, and develop quality metrics for hepatology.
AI-powered liver disease diagnosis
Liver biopsy is the gold standard for many chronic liver diseases (CLD), including liver fibrosis following chronic hepatitis, non-alcoholic fatty liver disease (NAFLD), cholangitis, and liver tumors or cysts.
Its invasiveness, particularly in patients already at high clinical risk for bleeding complications, and its complexity and cost prevent its use as a routine screening technique. Similarly, ultrasound imaging is useful in detecting and classifying many liver disorders but requires skilled operators and interpretation.
In this context, AI in hepatology offers non-invasive diagnosis and extends expert skill sets over multiple healthcare centers. ML and DL use algorithms that reflect or help to evolve the currently accepted guidelines for diagnosis, refined by application over large imaging and clinical datasets.
Radiomics refers to the computerized processing of imaging data to generate and classify a large amount of information, producing a prediction. Being so new, however, radiomics still awaits standardized protocols and criteria.
AI in liver disease is exemplified by the radiomics fibrosis index (RFI). This new DL-based model uses enhanced magnetic resonance imaging (MRI) to predict the stage of liver fibrosis. It exceeds the performance of other non-invasive clinical tests, such as the AST: platelet ratio and fibrosis-4 (FIB-4) index, avoiding liver biopsies in many cases.
For instance, liver fibrosis and CLD are common sequelae in chronic hepatitis B and chronic hepatitis C. Both these outcomes also pose independent risks for hepatocellular carcinoma (HCC). Diagnostic liver biopsies are impractical for screening all patients with chronic hepatitis, but ML-based screening offers rich promise.
Similarly, NAFLD is among the top causes of cirrhosis and of non-alcoholic steatohepatitis (NASH). An ultrasound-based AI program detected over 90% of NAFLD correctly. AI can provide faster, more objective interpretations of NASH histology while reducing pathologist workload.
DL-based visualization improves the diagnosis and treatment of significant portal hypertension. ML can also predict and stratify survival odds for primary biliary disorders. Moreover, ML can and should be used to pick up significant liver disease as an incidental finding in routine radiology reports.
Especially important is the role of AI in predicting compensated cirrhosis at the primary care level, enabling earlier treatment. AI performs better than physicians when distinguishing alcohol-linked hepatitis from acute cholangitis using clinical information.
Again, ultrasound-based ML algorithms outperformed experienced radiologists, yielding comparable information to CT or MRI for risk prediction and stratification of early HCC in cirrhosis patients and to differentiate benign from malignant tumors.
Machine learning for treatment planning
Liver transplants are the definitive therapy in terminal liver disease. AI-based liver segmentation for liver transplants helps plan liver resection and identify donor-recipient mismatch, improving graft and patient survival.
While dynamic contrast-enhanced MRI is the tool of choice today, advances in hepatology will see radiogenomics in use to classify tumors and predict their gene expression and mutation profile as accurately as a pathologist with five years of experience. This reveals tumor biology and prognosticates treatment responses.
A Deep Learning Treatment Assessment (DELTA) Liver Fibrosis score helps monitor fibrosis treatment by predicting reduced fibrosis severity. CT-based radiomics can predict high-risk varices (HRV) in patients with compensated severe CLD, thus limiting unnecessary procedures.
DL-based automatic segmentation of liver tumors overcomes the time and skill constraints of manual segmentation. This improves tumor load evaluation and treatment planning.
Predicting disease progression
DL models can help prioritize cirrhosis patients for liver transplants by predicting one-year mortality better than traditional analytics. They can predict post-transplant survival and complications.
AI tools can predict the outcomes for CLD, portal hypertension, esophageal varices, and the risk of acute or chronic liver failure (ACLF).
Automatic liver and tumor segmentation by AI accurately predicts tumor recurrence. DL can also accurately triage liver patients at high risk of short-term mortality.
The future of hepatology
AI-based advances in hepatology could impact the diagnosis, prognosis, and treatment of liver disease. AI uses massive datasets and machine analysis to counter bias of various types.
Image Credit: LALAKA/Shutterstock.com
Machine learning can help physicians interpret data better and faster, improve healthcare efficiency, and help patients improve their health. AI in hepatology also adds resources to practitioners in remote and low-resource locations, improving their training as well as facilitating better patient care.
AI can help identify therapeutic targets using clinical, molecular, and genetic data. By offering non-invasive severity assessment, ML can enhance NASH drug trial recruitment and promote new drug development. AI can predict drug trial results, promoting a faster and more accurate selection of candidate drugs.
The introduction of AI into clinical use is currently limited by the urgent need for careful validation and scrutiny of the algorithms in use, high-quality training, and testing datasets and techniques, and performance criteria, as well as for randomized clinical trials.
Since these algorithms operate automatically and on a large scale, “a flawed algorithm can cause harm to a large group of patients.” The potential for data theft and privacy breaches is also significant.
Once these issues are resolved, AI could drive precision and personalized medicine advances in hepatology.
References
Further Reading
Genetic study uncovers key drivers of metabolic syndrome with brain tissue links