IQAEvolNet: a novel unsupervised evolutionary image enhancement algorithm on chest X-ray scans

X-ray images are commonly used in medicine because they are inexpensive and non-invasive, but they often have noise issues caused by equipment or environmental factors. While there are many traditional image processing algorithms for enhancing X-ray images described in literature, there is a lack of machine learning-based methods due to the challenge of creating paired datasets of high-quality and low-quality images given the specific conditions under which these images are captured.

To address this issue, this paper proposes a new unsupervised evolutionary algorithm to improve chest X-ray images: Image Quality Assessment-based Evolutionary Network (IQAEvolNet).

The proposed approach involves training a convolutional neural network model using multiple image enhancement algorithms as references to establish an initial set of solutions. Then, an evolutionary algorithm is applied to refine these initial solutions, incorporating two image enhancement metrics, entropy and the Natural Image Quality Evaluator (NIQE), along with the Structural Similarity Index Measure, as fitness indicators.

The proposed algorithm achieved a lower NIQE score of 4.05 compared to 4.24 and a faster processing time of 2.95 milliseconds compared to 0.195 seconds, in relation to the state-of-the-art algorithm with the best NIQE and entropy.

It surpasses state-of-the-art algorithms in both NIQE score and processing time.