Deep Perceptual Enhancement for Medical Image Analysis

TL;DR

The paper addresses perceptual degradation in medical images caused by acquisition constraints and proposes a deep perceptual enhancement framework using a fully convolutional encoder–decoder with residual blocks and a contextual gating mechanism. It optimizes with a multi-term loss combining reconstruction, regularized feature loss on a pre-trained network, and an adversarial guidance to yield perceptually plausible results. Evaluations across Radiology, Dermatology, and Microscopy show consistent PSNR gains of about 5–7 dB and DeltaE reductions of about 4–6, along with improved expert preference and CAD performance, demonstrating real-world applicability. The work emphasizes robust generalization to real-world data and outlines future directions such as federated learning and 3D/video extensions to further enhance robustness and impact.

Abstract

Due to numerous hardware shortcomings, medical image acquisition devices are susceptible to producing low-quality (i.e., low contrast, inappropriate brightness, noisy, etc.) images. Regrettably, perceptually degraded images directly impact the diagnosis process and make the decision-making manoeuvre of medical practitioners notably complicated. This study proposes to enhance such low-quality images by incorporating end-to-end learning strategies for accelerating medical image analysis tasks. To the best concern, this is the first work in medical imaging which comprehensively tackles perceptual enhancement, including contrast correction, luminance correction, denoising, etc., with a fully convolutional deep network. The proposed network leverages residual blocks and a residual gating mechanism for diminishing visual artefacts and is guided by a multi-term objective function to perceive the perceptually plausible enhanced images. The practicability of the deep medical image enhancement method has been extensively investigated with sophisticated experiments. The experimental outcomes illustrate that the proposed method could outperform the existing enhancement methods for different medical image modalities by 5.00 to 7.00 dB in peak signal-to-noise ratio (PSNR) metrics and 4.00 to 6.00 in DeltaE metrics. Additionally, the proposed method can drastically improve the medical image analysis tasks' performance and reveal the potentiality of such an enhancement method in real-world applications. Code Available: https://github.com/sharif-apu/DPE_JBHI

Figures (11)

• Figure 1: The overview of the proposed deep perceptual enhancement method. The proposed network incorporates residual blocks and features a gating mechanism in an encoder-decoder like structure to reduce visual artefacts. Also, the proposed deep network has guided by the multi-term objective function, which aims to enhance the perceptual quality of low-quality images.
• Figure 2: Pair-wise image generation obtained by the proposed method. In every pair, top: reference image, and bottom: simulated low-quality image.
• Figure 3: Visualization of the training phase. The proposed method has trained until the model converges with the given images.
• Figure 4: Qualitative evaluation of proposed method and existing image enhancement methods. The proposed method illustrates consistency over every comparing medical image category. In every category, left to right: input (low-quality) images, results of LIME guo2016lime, DUAL zhang2019dual, RetnixNet wei2018deep, DRAN sharif2020learning, Proposed method and, ground truth images. (a) Radiology Images. (b) Dermatology Image. (c) Microscopic Images.
• Figure 5: Summary of expert preference. The medical image experts preferred the proposed method over comparing methods. (a) Proposed method. (b) DRAN sharif2020learning (c) LIME guo2016lime. (d) DUAL zhang2019dual. (e) RetnixNet wei2018deep. (f) Low-quality medical image.