Adaptive Distribution Learning With Statistical Hypothesis Testing for COVID-19 CT Scan Classification

Guan-Lin Chen, Chih-Chung Hsu, Mei-Hsuan Wu; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, 2021, pp. 471-479

With the massive damage in the world caused by Coronavirus Disease 2019 SARS-CoV-2 (COVID-19), many related research topics have been proposed in the past two years. The Chest Computed Tomography (CT) scan is the most valuable materials to diagnose the COVID-19 symptoms. However, most schemes for COVID-19 classification of Chest CT scan are based on single slice-level schemes, implying that the most critical CT slice should be selected from the original CT volume manually. In this paper, a statistical hypothesis test is adopted to the deep neural network to learn the implicit representation of CT slices. Specifically, we propose an Adaptive Distribution Learning with Statistical hypothesis Testing (ADLeaST) for COVID-19 CT scan classification can be used to judge the importance of each slice in CT scan and followed by adopting the nonparametric statistics method, Wilcoxon signed-rank test, to make predicted result explainable and stable. In this way, the impact of out-of-distribution (OOD) samples can be significantly reduced. Meanwhile, a self-attention mechanism without statistical analysis is also introduced into the backbone network to learn the importance of the slices explicitly. The extensive experiments show that both the proposed schemes are stable and superior. Our experiments also demonstrated that the proposed ADLeaST significantly outperforms the state-of-the-art methods.