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

Abstract

We introduce a lightweight model based on Mask R-CNN with ResNet18 and ResNet34 backbone models that segments lesions and predicts COVID-19 from chest CT scans in a single shot. The model requires a small dataset to train: 650 images for the segmentation branch and 3000 for the classification branch, and it is evaluated on 21292 images to achieve a 42.45% average precision (main MS COCO criterion) on the segmentation test split (100 images), 93.00% COVID-19 sensitivity and F1-score of 96.76% on the classification test split (21192 images) across 3 classes: COVID-19, Common Pneumonia and Control/Negative. The full source code, models and pretrained weights are available on https://github.com/AlexTS1980/COVID-Single-Shot-Model.

One Shot Model For The Prediction of COVID-19 and Lesions Segmentation In Chest CT Scans Through The Affinity Among Lesion Mask Features

“…This functionality was extensively used to train COVID-CT-Mask-Net [TS20a] and Single Shot Model (SSM) [TS20c]. These two models include several extensions of Mask R-CNN functionality for the global (image-level) prediction:…”

“…RoI branches [TS20c]. Augmentation of RoI to include the third parallel branch (in addition to detection and segmentation branches in Mask R-CNN): classification branch with the architecture identical to the detection branch.…”

“…We augment the RoI architecture with the fourth parallel branch that has the same architecture as Mask R-CNN mask branch, see Figure 2. Mechanism of the adaptation of this layer is the same as of the classification branch [TS20c]: the weights are copied from the mask branch if the segmentation mask loss improves. No other loss is computed, see Equation 8.…”

“…Often a combination of segmentation and classification helps with deriving saliency maps that increase the explainability of the model’s results, such as in [OPY20, WGM + 20]. For the purpose of explaining the classification results models based on Faster R-CNN [RHGS15] and Mask R-CNN [HGDG17] like COVID-CT-Mask-Net [TS20a] and Single Shot Model for COVID-19 prediction (SSM) [TS20c] demonstrate strong performance, due to Mask R-CNN’s ability to detect and segment instances of regions of interest (RoIs), i.e. the ability to understand the data at an object level.…”

One Shot Model For COVID-19 Classification and Lesions Segmentation In Chest CT Scans Using LSTM With Attention Mechanism

“…We refer to this as segmentation branch. RoI was augmented in [14, 15] to include a classification branch that also has two parallel branches, box and mask: box for the prediction of boxes and mask to extract mask features. Their architecture is identical to the segmentation branch.…”