ESPE Abstracts

Purpose: Turner Syndrome (TS) is a common chromosomal disorder in females, affecting approximately 3% of female conceptions. At present, the diagnosis of TS largely relies on karyotype testing and related technologies. The issuance of such tests depends on the experience of doctors and the analysis of reports takes more than a week, which leads to a high misdiagnosis rate for TS. Therefore, our objective is to facilitate swift and accurate identification of TS patients through the analysis of hand X-rays, thus enabling early intervention.

Methods: The study participants consisted of female individuals who were admitted to the hospital during the period from January 1, 2019, to December 31, 2022, due to their presentation of short stature. Each participant underwent comprehensive chromosome karyotype testing as part of their diagnostic evaluation. Our study introduces a patch-based deep learning decision fusion model to aid in TS diagnosis. The model comprises two main steps: first, a pre-trained hand landmark model annotates hand key points, and we utilize patches extracted according to key points to train the CNN encoders. Second, a decision fusion model combines patch-based classifications for a final diagnosis.

Results: Results show the model outperforms traditional methods, the best combination achieves an accuracy of 93.8% and a F1 score of 0.900. Compared to the traditional image-based CNN classification method, this combined model demonstrated a 14.95% improvement in accuracy, a 5.83% enhancement in recall, and an 18.58% increase in the F1 score. The findings of this investigation indicate that the distal segments of the second and third metacarpals, along with their corresponding proximal phalangeal regions, play a crucial role in the classification process.

Conclusion: Our Analysis identifies key features at bone junctions crucial for TS diagnosis, enhancing clinical understanding and aiding doctors in TS diagnosis. And approach of our study not only surpasses CNN-based methods but also tackles limited medical datasets of key points. research contributes to advancing clinical medicine and provides doctors with additional avenues for discovering valuable insights in TS diagnosis.