Developed AI/ML algorithm to detect ‘cotton wool’ infection Spots in retina images for a Leading Healthcare Technology Provider in the US
Overview
The client is one of the leading healthcare technology providers in the US, offering technology that empowers care delivery solutions for ophthalmic hospitals.
Download Case Study
Challenges
Detecting infection in the retina images, ‘cotton wool’ spots
Transforming and automating the entire manual diagnosis process
Calculating the spread of the infection and identifying the severity of the problem
Improve the accuracy of diagnosis
Solution
ACL Digital partnered with the leading healthcare technology player to develop an end application that would reduce manual efforts, speed up the examination process, and improve the accuracy of the process. As a preferred and reliable technology partner, below are some of the solutions that we provided:
- Model development:
- Pre-processing of the images
- Acquired relevant image dataset, which covers decent visual details of the infection.
- Performed annotation of the individual classes using the makesense.ai tool, which provides polygonal annotation in PVOC annotation format
- Deep learning training based Mask-RCNN model:
- It is a fully convolutional model for real-time instance segmentation, based on ResNet-101 with FPN (Feature Pyramid Networks)
- Trained the Mask-RCNN model using the annotated dataset to achieve higher accuracy and response time
- Model deployment:
- The trained model has been deployed on relevant Edge processing capable platform
Outcomes
- The deep-learning algorithm can be extended to simultaneously identify multiple retinal abnormalities, helping practitioners improve the early diagnosis of retinal diseases in underdeveloped areas, thus addressing the triple mandates of care, viz., - accessibility, affordability, and availability
- This automated detection system can reduce manual effort, and primary eye care services can be provided in remote areas to overcome the scarcity of doctors
- Accomplished a 95 percent decrease in exam time versus ophthalmologists working alone. The model helped lower exam time by 75 percent when combined with an ophthalmologist