Real-Time Abdominal Organs Classification in Ultrasound Using Deep Learning

Abstract

Manual identification of abdominal organs in ultrasound images is a time-consuming process that demands significant clinical expertise. Diagnostic accuracy often depends heavily on the operator’s experience, familiarity with organ-specific characteristics, and variations among patients, which can lead to inconsistent image interpretation. These challenges highlight the need for an automated and real-time system capable of assisting clinicians by reducing cognitive workload, improving reproducibility, and enhancing overall workflow efficiency.

To address this problem, we developed an automated abdominal organ classification system using YOLOv8, a state-of-the-art deep learning model optimized for real-time object detection and classification tasks. YOLOv8 was selected due to its high processing speed, advanced feature extraction capabilities, and strong performance in medical imaging applications. The model was trained on a labeled dataset of abdominal ultrasound images, with extensive data augmentation techniques applied to mitigate limitations in dataset size and improve the model’s generalization across diverse imaging conditions.

The performance of the proposed system was evaluated using standard data science metrics, including accuracy, precision, recall, and F1-score. The trained model achieved an accuracy of 94.44%, demonstrating effective classification of abdominal organs. Precision and recall values of 95.01% and 94.44%, respectively, indicate strong reliability in minimizing false positives while correctly identifying relevant organ structures. The resulting F1-score of 94.72% further confirms the model’s balanced performance across both detection sensitivity and classification accuracy.

Overall, the results demonstrate that YOLOv8 presents a feasible and highly effective solution for real-time abdominal organ classification in ultrasound imaging. This system shows strong potential for supporting clinical decision-making, reducing diagnostic variability, and improving the accuracy and efficiency of ultrasound-based examinations.