Executive Summary : | Glaucoma is popularly called the “Silent Thief of Vision”. It is one of the eye conditions that threaten irreversible retinal disorders. As per a recent report, the primary cause of Glaucoma is the damage of ganglion cells and rises Inter-Ocular Pressure (IOP). Primary open-angle glaucoma (POAG) affects about 57.5 million people worldwide. World Health Organization estimated that it has already affected 76 million people by 2020, and this figure may well rise to 111.8 million by 2040. Apart from IOP, Glaucoma may also be triggered by other factors, such as family history, gender, age, race, and even personality. Glaucoma can be detected by identifying structural changes in the Optic Disc (OD) and Optic Cup (OC) to estimate parameters such as Cup to Disc Ratio (CDR), Peripherical Artery (PPA), Neuro Retinal Rim (NRR), etc. However, detecting these parameters is a challenge due to uneven intensity distributions over the retinal surface of input images. But with the advent of Artificial Intelligence in the health sector, many different algorithms including Convolutional Neural Networks (CNN) have been proposed. (1) Although these methods have had a significant effect, they still suffer from many unavoidable constraints such as speed, memory size, etc. (2) Further, it is difficult to adjust the weights of CNN to focus more on the regions of interest. And finally, (3) It requires extensive training and makes the system increasingly complex. To circumvent these gaps, this proposal offers a low-cost economical multivariate feature driven by an AI-based health monitoring system for detecting early signs of Glaucoma. This proposal is mainly divided into two parts, the Software, and the Hardware. In the software part, an unsupervised boosting mechanism known as spiking convolution modules are introduced into the AI model to detect the complex patterns of input data with minimum resources. This concept of implementation of spiking convolution in the field of ophthalmology is novel and shall be the first of its kind methodology to implement on ophthalmic images to detect structural changes in the retina. (1) The proposed model includes exhibitory and inhibitory neurons to generate spike and synapse responses, enabling the proposed model to be at least four times faster than CNN with lesser memory consumption. (2) This multivariate feature-driven AI-based health monitoring system contains self-attention modules that increase the repeatability and reproducibility of the algorithm and effectively detect structural changes in the retina. In the hardware section, this AI model is implemented in the FPGA board using Xilinx Platform Studio in Verilog language. The prototype robust device will be installed in the collaborating hospital, Sankar Foundation Eye Hospital, for validation by Dr. Raveendra and his team of doctors. Finally, the prototype device will be equipped with a reasonably good configuration PC for providing detection of Glaucoma for rural people. |
Co-PI: | Dr. Bhaskara rao jammu, Gayatri Vidya Parishad College Of Engineering, Visakhapatnam, Andhra Pradesh-530048, Dr. Raveendra Tammineni, Sankar Foundation Eye Hospital, Visakhapatnam, Andhra Pradesh-530047 |