BONSAI Deep Learning Detects Pediatric Papilledema

The brain and optic nerve study with artificial intelligence (BONSAI) deep learning system (DLS) can distinguish papilledema from normal optic discs and from other optic disc abnormalities in children just as well as it can for adults, according to research presented at the 49th annual North-American Neuro-Ophthalmology Society (NANOS) meeting held in Orlando, FL, March 11-16, 2023. This approach could assist in diagnosing children in non-ophthalmic settings where a fundus camera is available, researchers report. 

The researchers reviewed the BONSAI deep learning system’s response to 898 fundus photographs (N=447; 61.5% girls; mean age, 10.33 years). The participants included 231 patients younger than 10 years, and 216 patients between 11 and 16 years. Findings included 558 discs normal, 254 instances of papilledema, and 86 other disc abnormalities

The researchers explained that multiclass accuracy of the BONSAI deep learning system was 89.6% and that it successfully distinguishes “normal” from “abnormal” optic discs with 87.3% sensitivity and 98.5% specificity. It also achieved a 98.0% sensitivity and 94.1% specificity distinguishing “papilledema” from the “normal and other” group. 

The performance of the BONSAI deep learning system was similar in both age groups, with overall accuracies of 89.2% in the patients younger than 10 years, and 90.1% in patients aged 11 to 16 years.

The BONSAI-DLS could be an effective diagnostic aid in non-ophthalmic pediatric settings, but real-life studies are necessary.

The BONSAI deep learning system for identifying papilledema from normal optic discs and other abnormalities using standard mydriatic ocular fundus photographs has recently been validated in an adult population, according to the presenters. In children, the presence of papilledema is sometimes a biomarker for severe neurologic disorders. Ocular fundus photographs are noninvasive and easily obtained from children. 

“In the pediatric population, the BONSAI-DLS displayed high sensitivity and specificity in differentiating papilledema from other optic disc abnormalities and normal optic discs,” the poster shows. “Performance was comparable across age, ethnicities, and study centers. The BONSAI-DLS could be an effective diagnostic aid in non-ophthalmic pediatric settings, but real-life studies are necessary.”

References:

Lin MY, Najjar RP, Tang Z, et al. The BONSAI deep learning system can accurately identify pediatric papilledema on mydriatic ocular fundus photographs. Presented at: North American Neuro-Ophthalmology Society 49th Annual Meeting; March 11-16, 2023. Poster 338.