Key Challenges
- The healthcare client needed a digital engineering partner to address the key challenge of early diagnosis of the disease to assist clinical practice
- Furthermore, it needed a solution for the identification of Diabetic Retinopathy in diabetic patients, and the solution needed to differentiate between retinal fundus images and other images
- The current diagnosis was cost-intensive, error-prone, and based upon one or a few doctors’ opinions
Solution
- Sparity designed a fundus degradation model based on the retinal ophthalmoscope imaging system to simulate low-quality fundus images
- Employed Kaggle dataset that provides a vast collection of high-resolution fundus images taken under a variety of imaging conditions
- Based on the fundus degradation model, a clinical-oriented fundus correction network was developed to correct low-quality fundus images for clinical observation and analysis
- The clinical-oriented fundus correction network model is built in PyTorch and optimized via stochastic gradient descent (SGD)
- Using the RSA module, the degradation algorithm effectively corrects low-quality fundus images without losing retinal information, while the LQA module suppresses unwanted artifacts
- The degradation algorithm also demonstrates that the fundus correction method is advantageous for medical image analysis applications, such as retinal vessel segmentation and optic disc/cup detection
- The degradation algorithm aids in the detection of diabetic retinopathy (DR) and helps ophthalmologists in the identification of ocular diseases through the study of retinal fundus images
Benefits
- 30% reduction in cost while maximizing investment
- 70% improvement in patient outcomes by providing early detection and treatment
- Improved diagnosis accuracy limits or even reverses the trend that characterizes the diffusion of such diseases
- Automated grading has potential benefits such as increasing efficiency, reproducibility, and coverage of screening programs, thereby reducing barriers to access
- Machine learning leverages the power of many doctors to come up with a diagnosis, improving patient outcomes by providing early detection and treatment
