A New AI Tool Revolutionizes the Detection of Rare Gastrointestinal Diseases
German scientists from Ludwig Maximilian University, the Technical University of Berlin, and the Charité clinic have developed an innovative artificial intelligence (AI) tool capable of identifying rare gastrointestinal tract diseases. This method, which has already undergone successful trials, promises to significantly expand diagnostic capabilities.
AI has become a key assistant in medical diagnostics, especially in analyzing visual data. However, traditional models are primarily trained on data from common diseases, making it difficult to apply them for detecting rarer pathologies. Researchers led by Professor Frederik Klauschen from the LMU Institute of Pathology and Professor Klaus-Robert Müller from the Technical University of Berlin developed a new method capable of addressing this issue. In their model, AI requires training only on data about normal conditions and common diseases to subsequently identify signs of rarer pathologies.
An Approach Based on Identifying Deviations from the Norm
The foundation of the new tool lies in the concept of anomaly detection. By analyzing samples of healthy tissues and common diseases, the model learns to identify deviations without being directly trained on examples of rare cases. To create the model, the scientists collected two extensive datasets of microscopic images of gastrointestinal biopsies, including both normal results and common diseases such as chronic gastritis. These data constituted about 90% of the total information. The remaining 10% included over 50 different diseases, including rare types of cancer.
For training and testing the model, the researchers used 17 million images of histological slices from 5423 cases. By comparing various technical approaches, the team selected a model that could accurately identify rare pathologies of the stomach and intestines, including infrequently encountered types of primary or metastatic cancer. Additionally, the AI tool can indicate specific areas of tissues affected by pathology using heat maps, which helps doctors quickly assess the results.
Send your records — an Ichilov physician will reply within 1–2 days, free of charge.
Reducing the Burden on Diagnosticians
The AI system, capable of independently identifying normal and common deviations, significantly accelerates the diagnostic process. Even when confirmation by pathologists is necessary, doctors will be able to save time by relying on AI to analyze normal results and frequently encountered diseases. According to Professor Klauschen, AI automatically processes about a quarter or even a third of all cases, allowing doctors to focus on more complex diagnoses and effectively prioritize their work. This significantly reduces the risk of missed diagnoses, representing an important step forward in medical practice.
Thus, the new AI tool provides doctors with effective support, improving the accuracy and speed of diagnosis while reducing the burden on specialists. In the future, this technology will continue to evolve, offering even more opportunities to enhance the quality of medical care for patients with rare gastrointestinal diseases.