Language assisted learnable hyperdimensional computing framework for retinal disease classification

Retinal diseases are considered the leading cause of blindness worldwide. Many researchers have developed autonomous retinal screening methods using optical coherence tomography (OCT), as they can reveal retinal abnormalities in the early stages. However, most of these methods lack the capability of performing retinal disease classification tasks by analyzing the disease-specific clinical manifestations, and they screen the retina by relying on mathematical features, which might not be meaningful for clinicians. To overcome these issues, we present a novel language-assisted learnable hyperdimensional computing (HDC) framework. HDC is a computational paradigm that represents and processes data using high-dimensional vector spaces to enable efficient, brain-inspired learning and reasoning. By fusing the language features and HDC visual embeddings computed through the set of clinical prompts and OCT scans, respectively, the proposed framework, with one-time training, can robustly recognize different retinal diseases irrespective of the scanner specifications, vendor artifacts, and dataset variations. The proposed framework is thoroughly tested on four public datasets, where it outperformed state-of-the-art methods in various metrics across each dataset.