Automated FDG uptake/PET-CT fused scan diagnosis of various lymph node tumors using object detection AI techniques

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lymph nodes (LN) are critical for immune function and fluid balance, and their accurate status assessment is vital for effective cancer treatment.
• Current imaging techniques often lack the sensitivity to detect small lymph node metastases, potentially leading to suboptimal treatment strategies and increased patient risk.
• Artificial intelligence (AI) object detection offers a promising avenue for enhancing the sensitivity and accuracy of lymph node characterization in medical imaging.

Purpose of the Study:

• To develop and evaluate an AI-based method for detecting 13 lymph node classes across various body organs using fused PET-CT imaging.
• To create a novel, annotated dataset by combining and processing PET-CT images for lymph node analysis.
• To compare the performance of a modified YOLOv8 object detection model against other state-of-the-art one-stage architectures for lymph node detection.

Main Methods:

• A new dataset was constructed by fusing, denoising, and annotating PET-CT images from real-world patient data.
• Data preprocessing included splitting into training, validation, and testing sets with augmentation applied only to the training set, followed by 5-fold cross-validation.
• A modified YOLOv8 object detection model was developed, featuring kernel selection, optimized backbone layers, and hyperparameter tuning, and compared against YOLOv7, YOLOv8, YOLOv9, YOLOv10, YOLOv11, YOLOv12, and YOLONas.

Main Results:

• The proposed modified YOLOv8 method demonstrated superior performance compared to other evaluated architectures.
• Significant improvements were observed in precision (78%), recall (75%), mean average precision (mAP50) (81%), and Dice similarity coefficient (DSC) (76%).
• The AI approach effectively addresses the limitations of conventional imaging in detecting small lymph node metastases.

Conclusions:

• The developed AI object detection method significantly enhances the accuracy and sensitivity of lymph node classification from PET-CT images.
• This advancement holds potential for more precise cancer staging and personalized treatment planning, reducing risks of inadequate or overly aggressive therapies.
• The study highlights the efficacy of AI in overcoming current challenges in medical image analysis for oncology applications.