Deep Learning Model Can Predict Cardiopulmonary Disease in Retinal Images of Preemies

MONDAY, Jan. 26, 2026 (HealthDay News) — A deep learning model using retinal images obtained during retinopathy of prematurity (ROP) screening may be used to predict diagnosis of bronchopulmonary dysplasia (BPD) and pulmonary hypertension (PH), according to a study published online Jan. 22 in JAMA Ophthalmology.

Praveer Singh, Ph.D., from the University of Colorado School of Medicine in Aurora, and colleagues examined whether images obtained as part of ROP screening might contain features associated with BPD and PH in infants and whether a multimodal model integrating imaging features with demographic risk factors can outperform a model based on demographic risk alone. A deep learning model was used to study retinal images collected from patients enrolled in the Imaging and Informatics in Retinopathy of Prematurity study; 493 infants at risk for ROP were included in the analysis. A support vector machine model was trained using image features alone, demographics alone, or image features and demographics.

The researchers found that the multimodal model showed higher accuracy than the demographics-only model or imaging only for BPD (area under the curve, 0.82 versus 0.72 and 0.72). The multimodal model showed significantly higher accuracy than the demographics-only model but not higher than the imaging-only model for PH (area under the curve, 0.91 versus 0.68 and 0.91, respectively). When trained on images lacking clinical ROP signs, the results persisted.

“We consider these results to be hypothesis generating and should be more fully evaluated in larger, heterogeneous populations and in images from other camera systems and modalities,” the authors write.

Several authors disclosed ties to the biopharmaceutical and eye care technology industries.

Abstract/Full Text (subscription or payment may be required)