EffUnet-SpaGen: An Efficient and Spatial Generative Approach to Glaucoma Detection.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Ophthalmology

Background:

• Automated disease detection algorithms are increasingly focused on developing "slimmer" models.
• These models aim to reduce the need for extensive training datasets and accelerate recalibration for new data while maintaining high accuracy.
• Developing slimmer models is a significant research trend in medical imaging.

Purpose of the Study:

• To develop a novel, efficient, and accurate two-phase automated glaucoma detection algorithm.
• To identify and leverage geometric redundancies in fundus image data for improved glaucoma diagnosis.
• To create a model that is computationally efficient and easily adaptable to new datasets.

Main Methods:

• Development of a novel cup and disc segmentation algorithm, "EffUnet", featuring an efficient convolution block.
• Integration of "EffUnet" with an extended spatial generative approach, "SpaGen", for geometry modeling and classification.
• Demonstration of rapid model training via recalibration of the EffUnet layer only.

Main Results:

• The EffUnet algorithm achieved high accuracy in segmenting optic disc and cup boundaries.
• The combined "EffUnet-SpaGen" algorithm surpassed state-of-the-art glaucoma detection methods, achieving AUROC scores of 0.997 (ORIGA) and 0.969 (DRISHTI).
• The algorithm provides explainability by visualizing deformed optic rim areas, crucial for clinical implementation.

Conclusions:

• The "EffUnet-SpaGen" algorithm significantly reduces computational burden in glaucoma detection.
• The model demonstrates superior accuracy and efficiency compared to existing methods.
• The explainability feature enhances the potential for clinical adoption and implementation of automated glaucoma detection systems.