Deep learning models that can accurately and efficiently identify measurements of tumors, tissue volume, or other types of abnormalities are used in disease diagnosis automation. Researchers have now unveiled a new, resource-light model capable of detecting a wide range of common eye diseases.
|A new deep learning model aids in the automated detection of common eye disorders./Tohoku University|
A group of Tohoku University researchers has unveiled new deep learning (DL) model that can identify disease-related features from images of eyes. This ‘lightweight’ deep learning model can be trained with a small number of images, even those with a high degree of noise, and is resource-efficient, allowing it to be deployed on mobile devices.
On May 20, 2022, details were published in the journal Scientific Reports.
With many societies aging and limited medical personnel, disease self-monitoring and tele-screening based on the DL model are becoming more common. Deep learning algorithms, on the other hand, are generally task specific, identifying or detecting general objects such as humans, animals, or road signs.
Disease identification, on the other hand, necessitates precise measurement of tumors, tissue volume, or other types of abnormalities. To do so, a model must examine separate images and draw boundaries, a process known as segmentation. However, accurate prediction requires more computational power, making them difficult to deploy on mobile devices.
“When it comes to DL models, there is always a trade-off between accuracy, speed, and computational resources,” says Toru Nakazawa, co-author of the study and professor at Tohoku University’s Department of Ophthalmology. “Our developed model has higher segmentation accuracy and improved model training reproducibility, even with fewer parameters, making it more efficient and lightweight than other commercial softwares.”
To create the model, Professor Nakazawa, Associate Professor Parmanand Sharma, Dr Takahiro Ninomiya, and students from the Department of Ophthalmology collaborated with Professor Takayuki Okatani of Tohoku University’s Graduate School of Information Sciences.
They obtained measurements of the foveal avascular zone, a region with the fovea centralis at the center of the retina, using low-cost devices to improve glaucoma screening.
“Our model can also detect/segment optic discs and hemorrhages in fundus images with high precision,” Nakazawa added.
The group hopes to use the lightweight model to screen for other common eye disorders and diseases in the future.
Reference: DOI: 10.1038/s41598-022-12486-w