Clinician-Led Code-Free Deep Learning for Detecting Papilledema and Pseudopapilledema Using Optic Disc Imaging

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Differentiating papilledema from pseudopapilledema is crucial for timely diagnosis and avoiding unnecessary procedures.
• Optical coherence tomography (OCT) and its near-infrared reflectance (NIR) imaging offer detailed optic nerve head visualization.

Purpose of the Study:

• To evaluate the performance of automated machine learning (AutoML) platforms in distinguishing papilledema from normal optic discs and optic disc drusen (ODD).
• To assess the capability of AutoML models in grading the severity of papilledema using NIR OCT images.

Main Methods:

• A retrospective cohort study analyzed 813 NIR OCT images from 289 patients with normal discs, papilledema, or ODD.
• Three AutoML platforms (Amazon Rekognition, Medic Mind, Google Vertex) were tested for classification and severity grading tasks.
• Performance metrics included area under the curve (AUC), precision, recall, and F1 score.

Main Results:

• Amazon Rekognition achieved the highest performance, with an AUC of 0.90 and F1 score of 0.81 for distinguishing papilledema from normal/ODD.
• For grading papilledema severity, Amazon Rekognition also led with an AUC of 0.90 and F1 score of 0.79.
• Google Vertex and Medic Mind demonstrated slightly lower accuracy and higher misclassification rates compared to Amazon Rekognition.

Conclusions:

• AutoML platforms are feasible for classifying papilledema using NIR OCT imaging.
• These models demonstrate potential for an accessible and scalable solution for clinical diagnosis of papilledema.
• Further external validation is required to confirm the clinical utility of AutoML in papilledema assessment.