A hybrid approach of simultaneous segmentation and classification for medical image analysis

Medical image analysis is a crucial step required for accurate disease diagnosis, treatment planning, and condition monitoring. In recent years, the field has undergone a groundbreaking transformation due to the advancement in artificial intelligence (AI) and deep learning (DL). These cutting-edge developments have particularly revolutionized automated segmentation and classification tasks, making them more efficient and reliable. The simultaneous performance of segmentation and classification enables the AI model to identify and isolate regions of interest, thereby enhancing the accuracy of the classification process. Several research reports showed that this simultaneous optimization posed a significant challenge, particularly in medical image analysis, of capturing intricate details and complex structures and having mitigating issues, such as vanishing gradient, distracting information, and multi-scale context. Therefore, this study proposed a hybrid hierarchical approach, SSC (Simultaneous Segmentation and Classification), integrating image segmentation and classification tasks within a redesigned network architecture. The main contribution of the study was the combination of the UNet architecture, classification block (C \(_{block}\) ${}_{\mathit{block}}$ ), residual block, attention mechanism, and Atrous Spatial Pyramid Pooling (ASPP) block together to achieve the improvement of segmentation and classification. A thorough comparative analysis and performance evaluation on two benchmark datasets, namely COVID-19 Radiography (CoR) and Breast Ultrasound Images (BUSI), were also conducted with 5-fold cross-validation. The experimental results showed that the proposed SSC model outperformed the baseline models. For the CoR dataset, the segmentation accuracy, Dice, IoU, classification accuracy, Recall, and Precision increased by 1.16%, 1.68%, 3.11%, 0.61%, 1.11%, and 1.20%, respectively. Meanwhile, for the BUSI dataset, 3.98%, 1.09%, 8.21%, 0.20%, 4.72%, and 1.55% increments were observed in the segmentation accuracy, Dice, IoU, classification accuracy, Recall, and Precision, respectively. These results showed the proposed method’s efficiency and potential for advancing simultaneous segmentation and classification tasks in medical image analysis.