Artificial intelligence (AI) has been at the forefront of recent efforts to develop precision medicine. Particularly, researchers are harnessing convolutional neural network (CNN) architectures for various radiological tasks. AI based on novel structures hold promise for predicting treatment responses for numerous disease conditions. A case in point is diabetic macular edema (DME). A major complication of diabetic retinopathy (DRP), DME has been key cause of visual impairment among working-age adults.
Selecting Best Treatment Plans for Patients with Vision Loss
Anti-vascular endothelial growth factor (VEGF) agents have become popular in retinal pathology. Such agents hold promise as the first line of therapy in DME patients as well. However, a few limitations defeats the purpose. The agents are not effective in all patients, and often require multiple injections. This catapults the cost. Hence, the need to identify the correct set of patient populations is vast, unarguably. Researchers found that AI can help to know if anti-VEGF treatments will work for a patient, and with good specificity and sensitivity.
Novel CNN Architecture Helped
They found that a novel CNN architecture they developed took into account optical coherence tomography (OCT) images. Moreover, these images are the standard of care in retinal pathology. Strangely enough, the AI didn’t need time-series OCT images to fulfil the objective. Rather, obtaining OCT images in any single pre-treatment point in time suffices. Furthermore, the researchers trained the AI to look for those CNN-encoded features that correlate with anti-VEGF response.
The respondents of the pilot study comprised 127 patients, all of which were undergoing anti-VEGF agent therapy. Moreover, they all had three consecutive shots. They found that their AI did a good task in identifying the cohort who will benefit– specificity of 85% and a sensitivity of 80%.
Further, the researchers asserted that the approach could help in other eye diseases, notably in age-related macular degeneration.