Multiclass Semantic Segmentation of COVID-19 CT Scan Images using Deep Learning

Artificial Intelligence (AI) has strengthened the medical image analysis field. Deep Learning has many applications in the area of COVID-19 diagnosis, disease progress monitoring, treatment response monitoring and quantization of severity of patients. In this work, Multiclass semantic segmentation of CT Scan images has been performed using a transfer learning approach. The DSC (Dice Similarity Coefficient) of baseline UNET segmentation model has improved from 93.7% to 95.33%, 96.93%, and 97.22% in UNET model using MobileNetV2, DenseNet121 and InceptionResNetV2 as encoder respectively.