Standardized diagnosis of image-based diseases may be possible with federated learning

Federated learning, a method used to train machine learning models without sharing patient data, has the potential to standardize clinical diagnosis in retinopathy of prematurity (ROP), according to a study.

A deep learning (DL)-derived ROP vascular severity score (VSS) developed using federated learning found differences in the clinical diagnosis of plus disease, and overall levels of ROP severity between several institutions.

The deep learning model for plus disease classification was trained using only the clinical labels from retinal imaging from the neonatal intensive care units of 7 institutions which were labeled with the clinical diagnosis of plus disease (plus, pre-plus, no plus) and a reference standard diagnosis (RSD) provided by 3 image-based ROP graders and the clinical diagnosis.

A VSS was determined for each eye exam using the 3 class probabilities as well as an institutional VSS of averaged VSS values for high severity eyes at each exam for each institution.

Researchers found significant variations between institutions in the proportion of patients diagnosed with pre-plus disease.

Differences across institutions were observed in the institutional VSS and the level of vascular severity diagnosed as no plus. Researchers also observed a significant, inverse relationship between the institutional VSS and the mean gestational age (P = 0.049, adjusted R²=0.49).

Reference
Hanif A, Lu C, Chang K; Imaging and Informatics in Retinopathy of Prematurity Consortium. Federated learning for multi-center collaboration in ophthalmology: implications for clinical diagnosis and disease epidemiology. Ophthalmol Retina. 2022;S2468-6530(22)00104-X. doi: 10.1016/j.oret.2022.03.005. Epub ahead of print. PMID: 35304305.