How AI is Changing the Way We Diagnose Coeliac Disease

AI Matches Pathologists in Diagnosing Coeliac Disease: A Breakthrough in Diagnostic Technology

A breakthrough study conducted by researchers at the University of Cambridge has shown that a machine learning algorithm is capable of diagnosing coeliac disease with an accuracy that rivals experienced pathologists. The study, published in the New England Journal of Medicine, demonstrates that AI can analyze biopsy images and correctly identify whether or not an individual has coeliac disease in 97 out of 100 cases. This advancement has the potential to speed up the diagnosis of this autoimmune condition, reduce strain on healthcare systems, and improve the accuracy of diagnoses, particularly in regions with limited access to experienced pathologists.

The Promise of AI in Medical Diagnostics

Machine learning (ML) and artificial intelligence (AI) have been transforming many industries, and healthcare is no exception. One area where AI tools have been showing particular promise is in the field of diagnostic pathology. By automating the analysis of medical images, AI has the potential to reduce the pressure on pathologists, streamline the diagnostic process, and help diagnose diseases faster and more accurately. While much of the focus in this area has been on cancer detection, emerging research is now exploring AI’s capability to diagnose a wide range of other diseases—coeliac disease being one of them.

Coeliac disease is an autoimmune disorder triggered by the ingestion of gluten in genetically predisposed individuals. It causes inflammation in the small intestine and results in symptoms such as abdominal cramps, diarrhoea, skin rashes, fatigue, and weight loss. However, these symptoms are not always present in every individual, and they can often mimic other gastrointestinal disorders, making early diagnosis a challenge. Without a proper diagnosis, coeliac disease can lead to long-term health complications like osteoporosis, infertility, and an increased risk of certain cancers.

The Current Diagnostic Approach for Coeliac Disease

Currently, the gold standard for diagnosing coeliac disease is through a biopsy of the duodenum, a part of the small intestine. Pathologists examine the biopsy under a microscope to assess the damage to the villi, tiny hair-like structures that line the intestine and play a critical role in nutrient absorption. The Marsh-Oberhuber scale is used to classify the severity of the damage, with scores ranging from zero (no damage) to four (complete villi flattening). However, interpreting these biopsies can be highly subjective, with subtle changes in the tissue that can sometimes be overlooked or misinterpreted.

The Cambridge Research Team and Their AI Tool

The team of researchers at Cambridge, led by Professor Elizabeth Soilleux, developed a machine learning algorithm trained on over 3,400 biopsy images from four NHS hospitals. The dataset included images from biopsies obtained using different types of scanners from five different hospitals and imaging companies. This diverse data set was critical for ensuring that the AI model could perform effectively under various conditions, mimicking real-world scenarios where biopsy images may be processed and analyzed differently.

The AI was trained to classify the biopsy images and diagnose coeliac disease with the same precision as experienced pathologists. To test the tool’s effectiveness, the team evaluated it using an independent dataset of 650 images, none of which had been part of the initial training. The results were impressive—the AI model diagnosed coeliac disease correctly in more than 97 cases out of 100.

Results and Accuracy of the AI Algorithm

The sensitivity of the AI model was greater than 95%, meaning it accurately identified more than 95% of individuals with coeliac disease. Its specificity was nearly 98%, indicating that it also correctly identified individuals who did not have coeliac disease in almost 98 out of 100 cases. This remarkable accuracy makes the AI model highly reliable, with performance comparable to that of a skilled pathologist.

In contrast, previous research by the Cambridge team showed that even human pathologists sometimes struggle to reach consensus when diagnosing coeliac disease. In one study, pathologists disagreed on the diagnosis in more than 20% of cases, highlighting the potential for diagnostic error in this challenging area. The AI tool, however, provided a consistent and reliable diagnosis across all datasets, demonstrating its ability to improve diagnostic accuracy.

The Role of AI in Healthcare Systems

This breakthrough research is an important step toward reducing the burden on healthcare professionals and improving the speed of diagnosis. Currently, pathologists are overwhelmed by the sheer volume of biopsies they need to analyze, particularly in busy healthcare systems like the NHS. AI-powered tools can help alleviate some of this pressure by automating routine diagnostic tasks, allowing pathologists to focus on more complex or urgent cases.

In addition, AI’s ability to work across various imaging systems and settings makes it a particularly useful tool in resource-limited environments. In countries or regions where there are severe shortages of pathologists, AI could help bridge the gap, enabling faster and more accurate diagnoses of conditions like coeliac disease, which might otherwise be delayed or missed entirely.

AI and the Future of Coeliac Disease Diagnosis

This development also holds promise for improving the experience of patients with coeliac disease, who often face long delays before receiving an accurate diagnosis. As Keira Shepherd, Research Officer at Coeliac UK, pointed out, the process of diagnosis often requires patients to maintain a gluten-rich diet, which can cause painful symptoms. Faster, more accurate diagnoses could help mitigate these difficulties and improve patients’ quality of life.

The Cambridge team is now working to expand the algorithm’s application in real-world clinical settings. The next step is to test it on a much larger sample of clinical data to ensure its effectiveness and to prepare for its regulatory approval. Ultimately, the goal is for this AI tool to be integrated into NHS diagnostic workflows and potentially become a globally accessible resource for improving coeliac disease diagnosis.

Conclusion: The Path Ahead

The ability of AI to accurately diagnose coeliac disease offers a glimpse into the future of medical diagnostics. With its impressive accuracy and ability to work across a variety of clinical settings, AI could become a key tool in addressing the diagnostic challenges faced by healthcare systems worldwide. The work of Cambridge’s research team is a testament to the potential of machine learning to revolutionize pathology and improve patient outcomes. As AI continues to evolve, it could soon play an integral role in diagnosing a broad range of diseases, helping clinicians provide faster and more precise care to patients in need.

Learn more: https://ai.nejm.org/doi/10.1056/AIoa2400738