# Artificial intelligence for diagnosing rare bone diseases: a global survey of healthcare professionals

## TL;DR

This study surveyed healthcare professionals about using AI to diagnose rare bone diseases, finding strong interest but also concerns about errors and transparency.

## Contribution

A global survey revealing healthcare professionals' openness to AI in diagnosing rare bone diseases and highlighting key concerns for practical adoption.

## Key findings

- Most respondents are open to integrating image recognition AI into their diagnostic workflow for rare bone diseases.
- Imaging, especially X-rays, is considered very important for diagnosing rare bone diseases by the majority of participants.
- Concerns about AI errors and model interpretability were reported by nearly half of the participants.

## Abstract

Rare bone diseases (RBDs) are an important group of conditions characterized by abnormalities in bone and cartilage. Their large number, individual rarity, and heterogeneity make accurate and timely diagnosis challenging. Establishing correlations between genotype and phenotype (mainly via imaging) is critical for diagnosing RBDs. Image recognition artificial intelligence (AI) has the potential to significantly improve the diagnostic process by assisting healthcare providers to identify and differentiate imaging patterns associated with various RBDs. This survey study sought to assess the interest of various healthcare providers worldwide in utilizing an AI-based assistant tool for the differential diagnosis of RBDs.

Survey data were collected from March to September 2024. The survey was performed online and the link was disseminated via direct email, newsletters, and flyers at scientific talks and conferences.

We received 103 completed surveys, representing respondents from 27 different countries covering most global regions, but mostly from Europe, the United States, and Canada. The majority of the participants are physicians (n = 92, 89%) and primarily work at academic medical centers (n = 84, 81%). While each participant could select multiple specialties, the most frequent clinician types were medical geneticists, pediatricians, and endocrinologists, accounting for 71 (69%) of the respondents. Ninety-four (91%) of the respondents find imaging to be very or extremely important, and the majority (n = 84, 81%) consider X-rays to be the most important imaging modality. Although around half of the participants (n = 45) have concerns about AI-related errors and consider the explainability of AI algorithms to be very (42/103) or extremely (9/103) important, 81% of the respondents report that they are somewhat (n = 39) or extremely (n = 45) likely to consider integrating image recognition AI into their current diagnostic workflow.

Most survey participants are open to integrating image recognition AI into their RBD diagnostic workflow. However, concerns about AI-related errors, privacy, and model interpretability highlight the importance of transparent collaboration between developers and healthcare professionals throughout the development process to ensure that such technologies are clinically trustworthy and practically adoptable.

The online version contains supplementary material available at 10.1186/s13023-025-03875-1.