Deep Learning-Based Detection of Abdominal Diseases Using YOLOv9 Models and Advanced Preprocessing Techniques

Artificial intelligence has emerged as a transformative tool in medical imaging, enabling automated diagnosis and analysis across various domains. While significant advancements have been made in abdominal imaging, many studies struggle to achieve robust detection of diseases. The complexity and variability in abdominal structures present unique challenges for traditional machine learning models, necessitating the adoption of more advanced object detection frameworks. Motivated by these challenges, this study focuses on leveraging the YOLOv9 object detection architecture to enhance the identification of abdominal diseases using the TEKNOFEST 2022 Abdomen Dataset. Advanced preprocessing techniques, including CLAHE (Contrast Limited Adaptive Histogram Equalization) and Gaussian noise augmentation, were applied to improve image contrast and model robustness. The dataset was processed into YOLO-compatible formats, and multiple training configurations were evaluated using YOLOv9c and YOLOv9s variants. These configurations included variations in batch size, optimizer type (SGD and Adam), dropout rate, and frozen layers. Among the configurations tested, the YOLOv9s model with 32 batch size, SGD optimizer, and a 35% dropout rate demonstrated the best performance, achieving a Recall of 0.7698, Accuracy of 0.7698, and F1 Score of 0.8228. The highest mAP50 of 0.9385 was observed with the YOLOv9c model trained using the Adam optimizer and a 35% dropout rate. Confusion matrix analysis revealed strong detection capabilities for conditions like acute cholecystitis and abdominal aortic aneurysm. This study highlights the potential of YOLOv9 models in medical imaging and emphasizes the importance of high-resolution datasets and advanced feature extraction techniques for improving diagnostic accuracy in abdominal disease detection. These findings lay a foundation for the development of reliable and efficient AI-driven diagnostic tools.