Applying Artificial Intelligence to 3D Medical Images
Two studies demonstrate the value of AI for detecting retinal and neurological problems.
Artificial intelligence provides a promising solution for medical image interpretation and triage, and has shown promise in classifying 2D photographs of some common diseases. According to two studies published August 13 in Nature Medicine, artificial intelligence approaches now show potential for spotting retinal and neurological abnormalities in 3D medical images.
In one study, researchers at DeepMind, an artificial intelligence company based in London, U.K., and collaborators applied a novel deep learning architecture to a clinically heterogeneous set of 3D optical coherence tomography scans from patients referred to a major eye hospital. After training on only 14,884 scans, the architecture demonstrated performance in making a referral recommendation that reached or exceeded that of experts for a range of sight-threatening retinal diseases. According to the authors, this work removes previous barriers to wider clinical use without prohibitive training data requirements across multiple pathologies in a real-world setting.
In the other study, researchers at the Icahn School of Medicine in New York City used a 3D convolutional neural network architecture to screen head computed tomography (CT) images for neurological illnesses. Features were automatically learned from a clinical radiology dataset consisting of 37,236 head CTs. A randomized, double-blinded, prospective trial in a simulated clinical environment demonstrated that this approach effectively triages the radiology workflow and accelerates the time to diagnosis from minutes to seconds. According to the authors, the speed offered by this machine-learning framework could improve outcomes for patients with neurological illnesses.