A new artificial intelligence (AI) tool has demonstrated the ability to significantly enhance the diagnosis of Sjögren’s disease by quickly and accurately analyzing tissue samples, according to a European multicenter study.
The AI system achieved more than 87% accuracy in classifying the focus score, an important measure of immune cell infiltration in the salivary glands, and predicting Sjögren’s diagnosis using minor salivary gland biopsies. This method is an important development because current manual assessment of focus scores requires specialized knowledge and is often highly variable among pathologists.
The researchers noted that AI “can reliably classify Focus Score and Sjögren’s disease using only minor salivary gland biopsies,” addressing a significant challenge in classifying this chronic autoimmune disease.
the study, “Machine learning to classify focus score and Sjögren’s disease using digitized salivary gland biopsies: a retrospective cohort study” was published. lancet rheumatology.
Understand Sjögren Scoring and Focus Scoring
Sjögren’s disease is an autoimmune disease in which the immune system mistakenly attacks healthy tissue, primarily the glands that produce tears and saliva. This causes severe dry eyes and characteristic symptoms of the mouth.
Currently, definitive diagnosis relies heavily on the evaluation of minor salivary gland biopsies, one of the two main classification criteria.
Biopsies are used to determine a foci score, which measures the number of foci, or clusters of at least 50 inflammatory immune cells, within a predefined tissue area. Diagnostic criteria require at least one focus score indicating immune cell infiltration or a positive blood test for Sjögren’s-related antibodies, particularly anti-SSA antibodies.
However, concentration scores are difficult to assess.
- Limited expertise: Accurate scoring requires highly specialized expertise, which only a few centers have.
- High error rate: In about half of cases, expert reassessment changes the diagnosis, often reducing a score of 1 or higher to a score of less than 1.
- Variability: Factors such as biopsy quality, infiltration pattern, and rater experience may vary.
“Despite the importance of this criterion for accurate disease classification, evaluation of focus scores remains a challenge,” the researchers wrote.
A team of researchers from six European centers leveraged machine learning, a type of AI that learns from data to identify complex patterns, to automatically classify focus scores and predict cases of Sjögren’s using digitally scanned biopsy slides.
For this study, the team trained an AI model on biopsy slide images collected from 356 adults with Sjögren’s disease and compared the results to a control group of 189 people with non-Sjögren’s sicca symptoms.
The machine learning model was trained on 80% of the slides from five centers and then validated using a completely different set of slides from the sixth center to ensure the reliability and generalizability of the results.
Model performance was evaluated using the area under the curve (AUC). This metric measures the model’s ability to distinguish between two groups (values closer to 1.0 indicate better performance). The results were very promising.
The AUC for differentiating patients with a focus score of at least 1 from those with a score <1 was 0.88 in the validation group. Among the validation set, the AUC was 0.81 for samples from patients with anti-SSA antibodies and 0.91 for samples from patients without these antibodies.
When retrained to predict Sjögren’s diagnosis based only on biopsy samples, the AUC was 0.89. Importantly, the model showed strong performance in a difficult-to-diagnose subset of patients, namely those who tested negative for anti-SSA antibodies. In this group, the diagnostic AUC reached 0.92.
Identifying tissue patterns
To understand how the AI was making decisions, the team conducted further analysis to determine which tissue patterns within the biopsy were involved in the model’s predictions.
Expert pathologists interpreted these patterns and confirmed that the AI was classifying diseases primarily based on the expected feature: clumps of immune cells that form lesions.
Interestingly, the model also discovered new patterns that are not currently included in Sjögren’s classification criteria. It identified specific immune cells, CD8-positive T cells (cytotoxic T cells), that surround acinar epithelial cells responsible for producing saliva in salivary glands.
“The artificial intelligence tool trained in this study reliably fulfills the unmet clinical need of focal score grading in patients with Sjögren’s disease, especially those who are negative for anti-SSA autoantibodies,” the researchers concluded.
The research team stressed that although the results were impressive, “the machine learning model used in this study needs to be validated in trials before it can be approved to assist pathologists in grading focal scores to provide more accurate Sjögren’s disease classification.”
