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

We present a novel framework that integrates segmentation of lesion masks and prediction of COVID-19 in chest CT scans in one shot. In order to classify the whole input image, we introduce a type of associations among lesion mask features extracted from the scan slice that we refer to as affinities. First, we map mask features to the affinity space by training an affinity matrix. Next, we map them back into the feature space through a trainable affinity vector. Finally, this feature representation is used for the classification of the whole input scan slice. We achieve a 93.55% COVID-19 sensitivity, 96.93% common pneumonia sensitivity, 99.37% true negative rate and 97.37% F1-score on the test split of CNCB-NCOV dataset with 21192 chest CT scan slices. We also achieve a 0.4240 mean average precision on the lesion segmentation task. All source code, models and results are publicly available on https://github.com/AlexTS1980/COVID-Affinity-Model. (C) 2021 Published by Elsevier B.V.

Automated summary

A novel framework that integrates lesion mask segmentation and COVID-19 prediction achieved high sensitivity and accuracy on the test split of the CNCB-NCOV dataset by introducing the concept of affinities for classifying the whole input image.