AI predicts cervical lymph node involvement from CT scans

TITLE

AI predicts cervical lymph node involvement from CT scans

PAPER TITLE

Automatic detection of cervical lymph nodes in patients with oral squamous cell carcinoma using a deep learning technique: a preliminary study

AUTHORS

Yoshiko Ariji; Motoki Fukuda; Michihito Nozawa; Chiaki Kuwada; Mitsuo Goto; Kenichiro Ishibashi; Atsushi Nakayama; Yoshihiko Sugita; Toru Nagao; Eiichiro Ariji

SUMMARY

This preliminary study applies a deep learning object detection approach to contrast-enhanced CT images to automatically detect cervical lymph nodes in patients with oral squamous cell carcinoma. The authors trained a DetectNet model using manually annotated bounding boxes and evaluated performance on an independent test set. The model achieved high precision and moderate recall, with better recall for metastatic nodes at cervical levels IB and II, suggesting potential as a supportive tool to reduce missed nodal findings in busy clinical workflows.

KEY WORDS

Deep learning, Object detection, Cervical lymph node metastasis, Oral squamous cell carcinoma, Computed tomography

EXTRACTED DATA

Year: 2020

Modality: Computed Tomography (contrast-enhanced CT)

Dataset: 56 patients with oral squamous cell carcinoma (OSCC)

365 axial CT images (cropped to 900×900 px)

Dataset Split: Training: 260 images ; Validation: 60 images; Testing: 45 images

Network Architecture: DetectNet object detector (NVIDIA DIGITS v5.0; DetectNet described as GoogLeNet-based subnet)

Training Details: 1000 epochs; ADAM solver; base learning rate 0.0001

Hardware: GeForce GTX 1080 Ti (11 GB); Ubuntu 16.04.2

Evaluation/Thresholds: Intersection over Union (IoU) threshold: 0.6 for a positive detection

Metrics: Recall (sensitivity), Precision (PPV), F1-score, False positive (FP) rate per image; Recall by cervical level

Performance (Test set): Metastatic LN: Recall 73.0%, Precision 96.4%, F1-score 83.1%, FP rate 0.04/image; Non-metastatic LN: Recall 52.5%, Precision 100.0%, F1-score 68.9%, FP rate 0.00/image

Recall by level (metastatic): Level IB 78.3%, Level II 80.0%, Other levels 55.6%

Annotation Process:

AP - Professional Qty: 1 radiologist

AP - Supervisor Presence: Not specified

AP - Experience Level: Not specified

AP - Expertise Area: Oral and Maxillofacial Radiology

Tool or System: ImageJ for annotation;

Task: Object Detection (localization + classification: metastatic vs non-metastatic lymph nodes)

Project Objective: Automatically detect cervical lymph nodes/metastasis on CT to reduce oversight and support staging in OSCC

CLINICAL RELEVANCE

Cervical lymph node status is essential for staging and prognosis in oral squamous cell carcinoma. An AI system capable of automatically detecting lymph nodes on CT could act as a second reader, helping reduce oversight, improving consistency, and supporting treatment planning and follow-up.