Artificial intelligence-enhanced ultrasound imaging for thyroid nodule detection and malignancy classification: a...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Thyroid nodules are common, necessitating accurate diagnosis for effective treatment.
• Traditional ultrasound diagnosis relies on physician experience, leading to variability.
• Artificial intelligence (AI) in medical imaging can improve diagnostic accuracy and efficiency.

Purpose of the Study:

• To evaluate the You Only Look Once v. 11 (YOLOv11) model for thyroid nodule detection and classification.
• To assess YOLOv11's effectiveness in ultrasound image analysis.
• To support real-time clinical decision-making and improve diagnostic workflows.

Main Methods:

• Utilized the YOLOv11 model on 1,503 thyroid ultrasound images (742 benign, 778 malignant).
• Employed data augmentation and transfer learning for model optimization.
• Compared YOLOv11 performance against other YOLO variants and ResNet50.

Main Results:

• YOLOv11 demonstrated superior performance over other YOLO variants and ResNet50.
• At IoU 0.5, YOLOv11 achieved precision (P) of 0.841 and recall (R) of 0.823.
• YOLOv11 showed high P and R for both benign and malignant nodules, with performance varying by IoU threshold.

Conclusions:

• YOLOv11 is a powerful tool for thyroid nodule detection and malignancy classification.
• The model offers high precision and real-time performance, crucial for dynamic ultrasound.
• Future work will expand datasets and validate clinical utility in real-time settings.