This post is also available in: עברית (Hebrew)
Researchers from Tohoku University in Japan have developed a lightweight deep learning model called LWBNA_Unet for automatic segmentation and analysis of eye images to help doctors identify and treat different diseases. The DL model is 10 times lighter than the most popular model in biomedical image segmentation, the Unet, and can be trained with a small number of images, including those with high levels of noise.
The model examines photographs and segments them by drawing lines along edges to assess tumors, tissue volume and other anomalies in the body. According to Marktechpost.com, researchers claim that while tele-screening for illnesses and self-monitoring based on DL models are becoming more commonplace, deep learning algorithms are often task-specific, recognizing or detecting broad objects like people, animals, or traffic signs.
“Our developed model has better segmentation accuracy and enhanced model training reproducibility, even with fewer parameters – making it efficient and more lightweight when compared to other commercial software.” Said Toru Nakazawa, an ophthalmology professor at Tohoku University and co-author of the study.