YOLOv8 Based An Automated Segmentation And Classification Of Thyroid Nodules Using USG Images

Thyroid nodules are a common clinical observation that needs to be accurately classified and segmented in order to be managed and treated. The use of the sophisticated object detection model You Only Look Once (YOLO) version 8 to the issue of nodule of the thyroid identification and categorization is investigated in this work. We gathered and interpreted 701 thyroid ultrasound images for this study, identifying two different nodule types as benign (315 images) and malignant (344 images). To assess the model's performance thoroughly, the dataset was divided into training (70%), validation (20%), and test (10%) sets. After training YOLOv8 on these images, metrics including Precision (78.1%), Recall (81.9%), F1 score (80%), and mean average precision (mAPall) (88.1% mAP@50) are used, with Benign (91.8% mAP) and Malign (84.5% mAP. The findings show that YOLOv8, outperforms earlier model YOLOv5, achieves good accuracy in both detecting and categorizing thyroid nodules. The ability of the model to accurately distinguish between benign and malignant nodules is critical for clinical decision-making.