Advancing COVID-۱۹ Infection Diagnosis: Integrating Segmentation and Classification via Deep Multi-Task Learning Model for Lung CT Scan Images

COVID-۱۹ presents a formidable global health challenge, underscoring the need for precise diagnosis of lung infections, specifically COVID-۱۹, to guide effective interventions. In response, this study introduces a multi-task model seamlessly merging segmentation and classification tasks for COVID-۱۹ detection from CT scan images. To address task imbalance arising from image processing algorithms during pre-processing, a strategic fusion of these algorithms is proposed. The model employs a U-net-based encoder-decoder architecture for comprehensive diagnosis. Pre-processing techniques, including median filtering, mathematical morphology, and histogram equalization, enhance image quality. The encoder extracts features through skip connections, while the decoder utilizes downsampling and convolution for segmentation. Classification is performed using a multilayer perceptron, with dataset masks and labels ensuring task compatibility. The proposed model was evaluated on four datasets and the combination of median filter and open morphology yields a remarkable dice coefficient of ۸۸.۹۱ ± ۰.۰۱ in segmentation and ۰.۹۷ classification accuracy were highest results by using the fourth dataset. In conclusion, the proposed model fostering the evolution of medical image analysis techniques and promising enhanced diagnostic precision and reliability in healthcare domains.

Keywords:

Medical image analysis , Image enhancement , Multi-task learning , Covid-۱۹ diagnosis