Automatic evaluation of 3D volume infection rates: Application to the Covid-19 context

ABSTRACT

Covid-19 is a highly contagious respiratory syndrome, officially declared a global pandemic on 11 March 2020. Due to its rapid spread and the exponential increase in the number of infected and deceased patients, manual diagnosis in the healthcare sector is insufficient to manage each patient individually, even the assessment of lesions by clinicians is approximate. Moreover, to date, no end-to-end tool is proposed for automatic volumetric quantification of Covid lesions. Hence, in this paper we report the implementation of a complete chain for automatic assessment of the degree of Covid-19 lesions. It includes (i) preparation of the private database, (ii) image pre-processing, (iii) automatic segmentation based on U-NET and evaluation of its results by the usual metrics, (iv) 3D reconstruction and finally (v) volumetric quantification of Covid-19 lesions using the digitised images as input. For validation, the process is applied to our own private database that we have created for this purpose. The results obtained are very encouraging. The evaluation of the segmentation for the lung by the metrics DICE, IOU, Precision, Recall and Accuracy yielded respectively 0.81, 0.90, 0.93, 0.82 and 0.92. Similarly for lesions these values are 0.89, 0.93, 0.93, 0.81 and 0.93 respectively. © 2022 IEEE.