COVID-19 identification from volumetric chest CT scans using a progressively resized 3D-CNN incorporating segmentation, augmentation, and class-rebalancing

The novel COVID-19 is a global pandemic disease overgrowing worldwide.\nComputer-aided screening tools with greater sensitivity is imperative for\ndisease diagnosis and prognosis as early as possible. It also can be a helpful\ntool in triage for testing and clinical supervision of COVID-19 patients.\nHowever, designing such an automated tool from non-invasive radiographic images\nis challenging as many manually annotated datasets are not publicly available\nyet, which is the essential core requirement of supervised learning schemes.\nThis article proposes a 3D Convolutional Neural Network (CNN)-based\nclassification approach considering both the inter- and intra-slice spatial\nvoxel information. The proposed system is trained in an end-to-end manner on\nthe 3D patches from the whole volumetric CT images to enlarge the number of\ntraining samples, performing the ablation studies on patch size determination.\nWe integrate progressive resizing, segmentation, augmentations, and\nclass-rebalancing to our 3D network. The segmentation is a critical\nprerequisite step for COVID-19 diagnosis enabling the classifier to learn\nprominent lung features while excluding the outer lung regions of the CT scans.\nWe evaluate all the extensive experiments on a publicly available dataset,\nnamed MosMed, having binary- and multi-class chest CT image partitions. Our\nexperimental results are very encouraging, yielding areas under the ROC curve\nof 0.914 and 0.893 for the binary- and multi-class tasks, respectively,\napplying 5-fold cross-validations. Our method's promising results delegate it\nas a favorable aiding tool for clinical practitioners and radiologists to\nassess COVID-19.\n