Deep Learning Based Automated COVID-19 Classification from Computed Tomography Images

The paper represents a method of a Convolution Neural Networks (CNN) model for image classification with image preprocessing and hyperparameters tuning, aiming at increasing the predictive performance for COVID-19 diagnosis while avoiding deeper and thus more complex alternatives. Firstly, the CNN model includes four similar convolutional layers followed by a flattening and two dense layers. This work proposes a less complex solution based on simply classifying 2D slices of CT scans using a CNN model. Despite the simplicity in architecture, the proposed CNN model showed improved quantitative results exceeding state-of-the-arts on the dataset of images, in terms of the macro F1 score. The results were achieved on the original CT slices of the dataset. Secondly, the original dataset was processed via anatomy-relevant masking of slices, removing non-representative slices from the CT volume, and hyperparameters tuning. For slice processing, a fixed-sized rectangular area was used for cropping an anatomy-relevant region of interest in the images, and a threshold based on the number of white pixels in binarized slices was employed to remove non-representative slices from the 3D-CT scans. The CNN model with a learning rate schedule with exponential decay and slice flipping techniques was deployed on the processed slices. The proposed method was used to make predictions on the 2D slices. For final diagnosis at a patient level, majority voting was applied on the slices of each CT scan to make the diagnosis. The macro F1 score of the proposed method well exceeded the baseline approach and other alternatives' scores on the validation set as well as on a test partition of the previously unseen images from the COV19-CT-DB dataset partition.