Bag of Tricks of Hybrid Network for Covid-19 Detection of CT Scans

This paper presents a study using deep learning models to analyze lung Computed Tomography (CT) images. Traditionally used for this task, deep learning frameworks face compatibility issues due to the variances in CT image slice numbers and resolutions caused by the use of different machines. Typically, individual slices are predicted and combined to obtain the final result, but this approach lacks slice-wise feature learning and ultimately leads to decreased performance. To address this limitation, we propose a novel slice selection method for each CT dataset, effectively filtering out uncertain slices and enhancing the model's performance. Moreover, we introduce a spatial-slice feature learning technique that uses a conventional and efficient backbone model for slice feature training. We then extract one-dimensional data from the trained COVID and non-COVID classification models by employing a dedicated classification model. Leveraging these experimental steps, we integrate one-dimensional features with multiple slices for channel merging and employ a 2D convolutional neural network for classification. In addition to the aforementioned methods, we explore various high-performance classification models, ultimately achieving promising results.