Automatic No-Reference kidney tissue whole slide image quality assessment based on composite fusion models. (April 2023)

• Record Type: Journal Article
• Authors: Ouyang, Jiazi; Ma, Xuetao; Wu, Yongfei; Li, Ming; Wang, Chen; Zhou, Xiaoshuang; Gao, Petting
• Is Part Of: Biomedical signal processing and control. Volume 82(2023)
• Publication Date: 2023-04
• Subjects: Deep learning -- Image quality assessment -- Network fusion -- Kidney tissue section -- Whole slide image; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.104547
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Propose a NR-IQA fusion networks for kidney tissue whole slide images. Utilize global and local information to jointly assess the overall image quality. Use discrete labels as input and obtain continuous results outcome. Extensive experiments are performed to verify the effectiveness of the method. Abstract: The visual quality of digitized whole slide images (WSIs) for kidney tissue not only affects the diagnosis and subsequent treatment, but also has a decisive impact on the accuracy of multiclass segmentation, classification and object detection during computer intelligent analysis. Currently, pathologists usually assess image quality through eye screening, which greatly relies on the pathologist experience and brings about subjectivity and non-repeatability issues. In this paper, we develop a no-reference image quality assessment framework including a fused CNN classification module, a quality score conversion module and a comprehensive quality prediction module, which automatically classifies WSIs of kidney tissue into four quality levels: excellent, good, average, and poor, and calculates a rough quality score. The original image and the regions of interest are combined and fused to comprehensively evaluate the quality of a WSI through multiple factors instead of a simple deep learning network. Extensive experiments conducted on our in-house dataset confirm that our proposed framework obtains satisfactory results with an accuracy of 90.05%, surpassing the performance of the typical image quality assessment models, and achieves the level of junior pathologist. Therefore, our proposed method can be embedded into a computer assisted diagnosis system to help pathologists in analysis of histopathological images and judgment of reliability of the images. The source code and trained models will be available at https://github.com/kidneyPathology/WSIQA .