[Paper Review] ADC-Net: An Open-Source Deep Learning Network for Automated Dispersion Compensation in Optical Coherence Tomography

Shaiban Ahmed, David Le+8 more|arXiv (Cornell University)|Jan 29, 2022

Retinal Imaging and Analysis4 citations

TL;DR

ADC-Net is an open-source deep learning framework based on a redesigned U-Net architecture that automates dispersion compensation in optical coherence tomography (OCT) by predicting fully compensated B-scans from partially compensated inputs. It achieves optimal performance with five input channels, improving image quality as measured by PSNR and MS-SSIM, and enables robust, automated resolution enhancement in retinal OCT imaging.

ABSTRACT

Chromatic dispersion is a common problem to degrade the system resolution in optical coherence tomography (OCT). This study is to develop a deep learning network for automated dispersion compensation (ADC-Net) in OCT. The ADC-Net is based on a redesigned UNet architecture which employs an encoder-decoder pipeline. The input section encompasses partially compensated OCT B-scans with individual retinal layers optimized. Corresponding output is a fully compensated OCT B-scans with all retinal layers optimized. Two numeric parameters, i.e., peak signal to noise ratio (PSNR) and structural similarity index metric computed at multiple scales (MS-SSIM), were used for objective assessment of the ADC-Net performance. Comparative analysis of training models, including single, three, five, seven and nine input channels were implemented. The five-input channels implementation was observed as the optimal mode for ADC-Net training to achieve robust dispersion compensation in OCT

Motivation & Objective

• To address chromatic dispersion, a major source of resolution degradation in optical coherence tomography (OCT).
• To develop an automated, deep learning-based solution for dispersion compensation that eliminates manual tuning and subjective adjustment.
• To optimize image quality across all retinal layers by predicting fully compensated B-scans from partially compensated inputs.
• To evaluate and identify the optimal number of input channels for training the deep learning model.

Proposed method

• ADC-Net employs a modified U-Net encoder-decoder architecture to learn the mapping from partially compensated to fully compensated OCT B-scans.
• The input consists of multiple OCT B-scans with individual retinal layers already optimized for dispersion, enabling the network to learn residual compensation.
• The network is trained using a loss function that minimizes reconstruction error, with performance evaluated using PSNR and multi-scale structural similarity (MS-SSIM).
• Five different input configurations (1, 3, 5, 7, 9 channels) were tested to determine the optimal input dimension for robust performance.
• The model is trained end-to-end on real OCT data, with the goal of restoring full resolution across all retinal layers.
• The framework is released as open-source, enabling reproducibility and integration into clinical and research workflows.

Experimental results

Research questions

• RQ1 What is the optimal number of input channels for effective dispersion compensation in OCT using deep learning?
• RQ2 Can a deep learning model achieve consistent and objective dispersion compensation across diverse OCT B-scans without manual intervention?
• RQ3 How does the performance of the ADC-Net model compare to conventional methods in terms of image quality metrics like PSNR and MS-SSIM?
• RQ4 To what extent does the U-Net-based architecture improve the accuracy and robustness of automated dispersion compensation in OCT?

Key findings

• The five-input-channels configuration achieved the best performance in terms of PSNR and MS-SSIM, outperforming models with one, three, seven, or nine input channels.
• ADC-Net significantly improved image quality by restoring resolution across all retinal layers in OCT B-scans.
• The model demonstrated robustness and generalization across diverse OCT imaging data, indicating strong potential for clinical deployment.
• Objective metrics (PSNR and MS-SSIM) confirmed that the predicted fully compensated B-scans closely matched ground-truth quality, validating the model's effectiveness.
• The open-source release of ADC-Net enables broader adoption and integration into OCT imaging pipelines for research and clinical use.