Pascal Sève, Professeur de médecine interne - Chef du département de médecine interne de l'hôpital Croix-Rousse
Robin Jacquot, PhD Student, Reshape U1290
How can we enhance the etiological approach for patients with uveitis?
Background: The large number and heterogeneity of causes of uveitis make the etiological diagnosis a complex task. The clinician must consider all the information concerning the demographic, ophthalmological, and extra-ophthalmological features of the patient. Although diagnostic machine learning algorithms have been developed, providing a correct diagnosis in one-half to three-quarters of cases, they are not integrated into daily clinical practice and require further improvement.
Materials and methods: A neural network (multi-layer perceptron) was trained on a dataset of 874 incident cases of uveitis with unknown etiology and tested on 375 cases. Twenty-five causes of uveitis were selected, excluding cases diagnosed solely by ophthalmologists. The algorithm analyzed relevant demographic, ophthalmological, and clinical factors, along with four complementary exams. To assess the performance of the neural network, the gold standard was the etiological diagnosis established by a consensus of two uveitis experts following the classification criteria available in the literature.
Results: The algorithm’s most probable diagnosis matched the senior clinician diagnosis in 292 of 375 patients (77.8%, 95% CI: 77.4-78.0). It achieved 93% accuracy (95% CI: 92.8-93.1%) when considering the two most probable diagnoses. The algorithm performed well in diagnosing idiopathic uveitis (sensitivity of 81% and specificity of 82%). For more than three-quarters of etiologies, our algorithm demonstrated good diagnostic performance (sensitivity > 70% and specificity > 80%).
Conclusion: We have successfully developed an accurate algorithm for the clinical management of uveitis with undetermined etiology which can be considered as an aid in the etiological assessment process.
Keywords: uveitis; artificial intelligence; etiology; algorithm; diagnosis; neural network; machine learning.
