Orphanet J Rare Dis. 2025 Apr 17;20:186 

PMID: 40247315 DOI: 10.1186/s13023-025-03655-x 

Dominique P. Germain, David Gruson, Marie Malcles, Nicolas Garcelon

Highlights

This review explores current and potential applications of artificial intelligence (AI) in the field of rare diseases, using Fabry disease as a model condition. Based on a targeted literature review, the study highlights how AI can support both prospective screening and retrospective diagnosis, in addition to aiding the monitoring of organ involvement and the development of personalized treatment strategies. Linear modeling, natural language processing (NLP), image analysis, and multimodal approaches are emphasized as effective tools for accelerating diagnosis. 

Background

Rare diseases, though individually uncommon, collectively affect over 300 million people worldwide. The diagnostic journey is often long and complex—referred to as a "diagnostic odyssey"—and frequently results in delayed diagnosis, misdiagnosis, or lack of diagnosis altogether. Fabry disease, a lysosomal storage disorder, is one such condition where early diagnosis and intervention can prevent irreversible organ damage. This context underscores the value of AI-driven diagnostic approaches. 

Material and Methods

A PubMed search was conducted using the keywords "AI and Fabry", "ML and Fabry", "DL and Fabry", and "AI and rare diseases". Twenty peer-reviewed articles published between 2021 and 2023 were selected for inclusion. The selected studies applied machine learning (ML) and deep learning (DL) techniques to domains such as electronic health record (EHR) mining, image-based diagnostics (cardiac, brain, ocular, dermatologic, auditory), and genomic/proteomic analysis in relation to Fabry disease. 

Findings

EHR-based screening: NLP algorithms successfully extracted Fabry-associated symptoms from clinical notes, demonstrating effectiveness in identifying rare metabolic diseases. 

Facial recognition: AI models like DeepGestalt differentiated facial morphologies characteristic of Fabry patients from other syndromes. 

Cardiac analysis: ML models interpreted EKG and cardiac MRI data to distinguish Fabry disease from other cardiomyopathies and predict disease progression. 

Brain aging models: DL tools applied to brain MRI estimated accelerated aging profiles in Fabry patients. 

Retinal and dermatologic imaging: DL models showed promise in gender classification and rare disease discrimination using fundus images; similar methods are under development for identifying eye and skin manifestations in Fabry disease.

Renal pathology: CNN-based image classifiers predicted podocyte damage from electron microscopy data, aiding renal monitoring in Fabry patients. 

Conclusion

AI technologies hold transformative potential for the diagnosis, monitoring, and management of rare diseases. Fabry disease serves as a compelling case demonstrating the clinical feasibility of AI-based tools. Ethical governance of clinical data, patient privacy, and physician oversight remain essential. Regulatory frameworks like the European Union’s AI Act could facilitate safe integration of AI into healthcare systems. Widespread implementation of multimodal AI models may enable earlier, more accurate, and more personalized care for rare disease patients. 

Keywords: Artificial intelligence; Deep learning; Fabry disease; Machine learning; Rare diseases.