Camera-Agnostic OCT Generation using a Causal Inference-based Quantum Computing Transfer Model (CIQCT)

Abstract

Purpose: Extensive research has been conducted on deep learning (DL) algorithms for automated generation of multimodal ophthalmic imaging and disease detection systems using traditional convolutional neural networks (CNN). However, the performance of these systems is a function of the specific mode of photo acquisition- use of images from one type of camera has shown inconsistent performance with another. We propose a camera-agnostic casual inference-based quantum computing deep neural network (CI-QNN) algorithm for reliable, high-quality generation of normal and pathological OCT photos. The model aims to improve on the classical CNN generated images with enhanced quantum image acquisition, potentially improving the diagnostics and management of retinal pathology.

Methods: The CI-QNN will be developed based on the methods-based AI “cofe-Net” for image enhancement and normalization. The model will be trained with images from various national image datasets (DRCR.net, EyePACS). External validation will be performed with fundus and OCT photos from 213 participants from the US and Bangladesh. Metrics quantifying the structural similarity between generated and real images based on luminance, contrast, and structure will be tabulated, including the average Structural Similarity Index Measure (SSIM), average Peak Signal-to-Noise Ratio (PSNR), and average Mean Squared Error (MSE) for the non-pathological and pathological images.

Results: The developed model produced a non-pathological OCT image with a SSIM of 70.9%, PSNR of 17.6023 dB, and average MSE of 0.017429. Work is currently ongoing to improve image generation accuracy with further training on datasets with various image sources. Both normal and pathological images will be generated throughout the process, with the plan to augment the algorithm for optimal image generation.

Conclusions: The CI-QCT can reliably create OCT images from fundus photos regardless of camera type. The model continues to expand its generative capabilities through quantum computing methods with anticipated creation of high-quality, reliable pathological OCT images soon. It can enhance disease monitoring and stratification for patients with barriers to multimodal imaging and aid in the preservation of vision and quality of life.