DeepLabV3+-Based Automated System for Classification and Segmentation of Fatty Liver Disease in Ultrasound Images

Fatty Liver Disease (FLD) is a growing global health issue, often progressing silently without overt symptoms until advanced stages. As a result, early detection is critical to prevent complications such as cirrhosis and hepatocellular carcinoma. Ultrasound imaging offers a non-invasive, affordable, and widely accessible diagnostic modality for liver disorders. However, manual interpretation of ultrasound scans is subject to inter-observer variability and diagnostic delays. To address this challenge, this study presents an automated system for the classification and segmentation of FLD using the DeepLabV3+ semantic segmentation model. Built on an encoder-decoder architecture with atrous spatial pyramid pooling, DeepLabV3+ efficiently captures multi-scale contextual features and sharp object boundaries, making it well-suited for medical image analysis.The model was trained and tested on a curated dataset comprising annotated ultrasound liver images representing both healthy and FLD conditions. Preprocessing included normalization, resizing, and augmentation to enhance model generalization. The system demonstrated robust performance with a classification accuracy of 96.3%, Dice coefficient of 0.91, Jaccard Index of 0.84, precision of 0.94, and recall of 0.88. Visual inspection of the segmentation results confirmed that the model could accurately delineate liver regions and fatty infiltration areas, which are vital for clinical decision-making.