Deep Learning-Based Detection of Malignant and Equivocal Cervical Lymph Nodes on CT Imaging Using a 2D U-Net++ Model

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Accurate identification of malignant and equivocal lymph nodes on computed tomography (CT) is challenging, especially in borderline cases.
• Artificial intelligence (AI) offers potential for enhancing diagnostic performance in radiology.

Purpose of the Study:

• To evaluate the performance of a 2D U-Net++ convolutional neural network for detecting malignant and equivocal lymph nodes on CT imaging.
• To assess the AI model's ability to support radiological decision-making in clinical practice.

Main Methods:

• Retrospective analysis of 79 CT examinations with 538 annotated lymph nodes (Node-RADS 3-5).
• Implementation and training of a 2D U-Net++ model using NIfTI data with data augmentation.
• Performance evaluation using metrics including accuracy, sensitivity, specificity, Dice Similarity Coefficient (DSC), and Intersection over Union (IoU).

Main Results:

• The AI model achieved high accuracy (99.7% on validation) and specificity (99.9%) in lymph node detection.
• Segmentation performance showed a Dice Similarity Coefficient (DSC) of 0.73 and Intersection over Union (IoU) of 0.61 on validation.
• The model demonstrated robust anatomical discrimination, correctly excluding mimics and showing limited false-negative and false-positive results.

Conclusions:

• The 2D U-Net++ model reliably detects malignant and equivocal lymph nodes on CT, offering high specificity and anatomical discrimination.
• The AI tool shows potential as a clinical decision-support system, particularly for highlighting borderline findings in routine CT scans.