Lightweight Model For The Prediction of COVID-19 Through The Detection And Segmentation of Lesions in Chest CT Scans

Abstract

We introduce a lightweight Mask R-CNN model that segments areas with the Ground Glass Opacity and Consolidation in chest CT scans. The model uses truncated ResNet18 and ResNet34 nets with a single layer of Feature Pyramid Network as a backbone net, thus substantially reducing the number of the parameters and the training time compared to similar solutions using deeper networks. Without any data balancing and manipulations, and using only a small fraction of the training data, COVID-CT-Mask-Net classification model with 6.12M total and 600K trainable parameters derived from Mask R-CNN, achieves 91.35% COVID-19 sensitivity, 91.63% Common Pneumonia sensitivity, 96.98% true negative rate and 93.95% overall accuracy on COVIDx-CT dataset (21191 images). We also present a thorough analysis of the regional features critical to the correct classification of the image. The full source code, models and pretrained weights are available on https://github.com/AlexTS1980/COVID-CT-Mask-Net.

Single-Shot Lightweight Model For The Detection of Lesions And The Prediction of COVID-19 From Chest CT Scans 

“…1. Reducing the model's depth from full (4 layers) to 3 and 2, while keeping a single FPN module (see [TS20c] on the matter of model truncation for this problem),…”

“…The segmentation model was pretrained for 50 epochs with Adam optimizer, learning rate of 1e − 5 and weight decay coefficient of 1e − 3. Important Mask R-CNN hyperparameters such as non-maximum threshold were the same as in [TS20c].…”

“…The batch of these predictions is converted into a vector of features that the classification module in COVID-CT-Mask-Net must learn. In [TS20b,TS20c] a number of modifications were presented, including truncated lightweight versions of the base model that achieve a similar accuracy of COVID-19 prediction and lesion segmentation. The main drawback of this approach is that it consists of two stages: first, a segmentation model (Mask R-CNN) is trained on the data with ground truth (gt) masks, then, a classification model is derived from it and trained on the data labeled at the image level (COVID-19, Common Pneumonia, Control).…”

Single-Shot Lightweight Model For The Detection of Lesions And The Prediction of COVID-19 From Chest CT Scans

“…For the 3-class (COVID19, CP, Normal) classification data we use the COVIDx-CT train/test/validation splits, [GWW20]. We use the same sample of 3000 images (1000/class) from the train split as in [TS20c], and also used the test and validation splits in full, see Table 1. The splits are consistent across classes and patients.…”

“…The batch of these predictions is converted into a vector of features that the classification module in COVID-CT-Mask-Net must learn. In [TS20b, TS20c] a number of modifications were presented, including truncated lightweight versions of the base model that achieve a similar accuracy of COVID-19 prediction and lesion segmentation. The main drawback of this approach is that it consists of two stages: first, a segmentation model (Mask R-CNN) is trained on the data with ground truth (gt) masks, then, a classification model is derived from it and trained on the data labeled at the image level (COVID-19, Common Pneumonia, Control).…”