Interpretable laryngeal tumor grading of histopathological images via depth domain adaptive network with integration gradient CAM and priori experience-guided attention. (March 2023)

• Record Type: Journal Article
• Authors: Huang, Pan; Zhou, Xiaoli; He, Peng; Feng, Peng; Tian, Sukun; Sun, Yuchun; Mercaldo, Francesco; Santone, Antonella; Qin, Jing; Xiao, Hualiang
• Is Part Of: Computers in biology and medicine. Volume 154(2023)
• Publication Date: 2023-03
• Subjects: Laryngeal cancer -- Tumor grading -- Attention mechanism -- Visually interpretable -- Domain adaptive; Medicine -- Data processing -- Periodicals; Biology -- Data processing -- Periodicals; 610.285
• DOI: 10.1016/j.compbiomed.2022.106447
• Languages: English
• ISSNs: 0010-4825

Abstract

Abstract: Tumor grading and interpretability of laryngeal cancer is a key yet challenging task in the clinical diagnosis, mainly because of the commonly used low-magnification pathological images lack fine cellular structure information and accurate localization, the diagnosis results of pathologists are different from those of attentional convolutional network -based methods, and the gradient-weighted class activation mapping method cannot be optimized to create the best visualization map. To address this problem, we propose an end-to-end depth domain adaptive network (DDANet) with integration gradient CAM and priori experience-guided attention to improve the tumor grading performance and interpretability by introducing the pathologist's a priori experience in high-magnification into the depth model. Specifically, a novel priori experience-guided attention (PE-GA) method is developed to solve the traditional unsupervised attention optimization problem. Besides, a novel integration gradient CAM is proposed to mitigate overfitting, information redundancies and low sparsity of the Grad-CAM graphs generated by the PE-GA method. Furthermore, we establish a set of quantitative evaluation metric systems for model visual interpretation. Extensive experimental results show that compared with the state-of-the-art methods, the average grading accuracy is increased to 88.43% (↑4.04%), the effective interpretable rate is increased to 52.73% (↑11.45%). Additionally, it effectively reduces the difference between CV-based method and pathology in diagnosis results. Importantly, the visualized interpretive maps are closer to the region of interest of concern by pathologists, and our model outperforms pathologists with different levels of experience. Highlights: We introduced pathologist the pathologist's a priori experience in high-magnification into low-magnification images via DDANet to narrow the difference in diagnostic results between pathologists and CV-based method. Besides, a quantitative evaluation metric system is established for the visual interpretation of laryngeal cancer pathology grading models. We propose a novel scale-invariant feature transform, which uses the K-nearest neighbor, row scanning, down sampling (SKRD) algorithms to improve the efficiency of pathologist labeling and tumor grading. A priori experience-guided attention algorithm is proposed to solve the problem of unsupervised optimization of traditional AMs, the ROI of DDANet is made more consistent with the ROI of pathologists. An integration gradient CAM algorithm is developed to have multiple supervised fits of the down-sampled pathologist mask maps, thus improving the accuracy of the DDANet visualization maps.