Automated Detection and Diagnosis of Pathological DRGs in PHN Patients Using Deep Learning and Magnetic Resonance

Huazhong University of Science and Technology

Status: Completed

Conditions: Deep Learning

Treatments: Other: YOLOv8

Study type: Observational

Funder types: Other

Identifiers

NCT06274502

deep learning

Details and patient eligibility

About

Here, this study aimed to develop an automated system for detecting and diagnosing lesion DRGs in PHN patients based on deep learning. This study retrospectively analyzed the DRG images of all patients with postherpetic neuralgia who underwent magnetic resonance neuroimaging examinations in our radiology department from January 2021 to February 2022. After image post-processing, the You Only Look Once (YOLO) version 8 was selected as the target algorithm model. Model performance was evaluated using metrics such as precision, recall, Average Precision, mean average precision and F1 score.

Full description

Our previous research has confirmed differences in macroscopic and microscopic aspects between the imaging of lesioned dorsal root ganglia (DRG) in patients with postherpetic neuralgia (PHN) and healthy controls. Additionally, our study revealed that while lesioned skin localization is a classic method in clinical practice, there is still a certain rate of discrepancy with the lesioned DRG observed in magnetic resonance imaging (MRI). This suggests the significant value of MRI in diagnosing lesioned DRG in PHN patients. For patients with zoster sine herpete neuralgia, it is even more crucial to identify the lesioned DRG through MRI. However, due to the small size and varied morphology of DRG lesions, diagnosing lesioned DRG through MRI requires specialized knowledge in neuroanatomy and imaging, posing a challenge to clinical practitioners. Identifying lesioned DRG rapidly and accurately is crucial for interventional therapy, as it serves as an essential treatment target for neuropathic pain.

The YOLO (You Only Look Once) series of algorithms are currently widely used single-stage real-time object detection algorithms, including YOLOv1-YOLOv8. Due to their extremely high detection speed, they enable real-time object detection. YOLOv5 and YOLOv8 are now extensively employed in various applications such as autonomous driving, video surveillance, and object tracking. Moreover, the YOLO series is increasingly being applied in the medical field, including tumor and joint capsule lesion detection, demonstrating good accuracy, recall rates, and detection efficiency. This study aims to utilize the YOLOv8 algorithm to develop a fast and accurate object detection model, simultaneously evaluating its performance. It seeks to validate the feasibility and effectiveness of detecting lesioned dorsal root ganglia (DRG) in real-time postherpetic neuralgia using this model, providing a basis for early diagnosis for clinical practitioners and enabling rapid and precise localization of lesioned DRG.

Enrollment: 41 patients

Sex: All

Ages: 18 to 85 years old

Volunteers: No Healthy Volunteers

Inclusion criteria

• Patients aged 18 years or older;
• Patients with herpes zoster who continue to experience pain for over one month after the crust forms over the skin lesions;
• Clear MRI images showing evident dorsal root ganglia (DRG) lesions.

Exclusion criteria

• Patients with severe systemic, metabolic, or neurological diseases that can lead to polyneuropathy, such as multiple myeloma, diabetes, or thyroid diseases;
• Patients with a history of psychiatric disorders, other chronic pain conditions, or substance abuse;
• Patients with a history of thoracic or back surgeries and a history of pain;
• Presence of artifacts in the imaging or unclear image display;
• Target images obscured by other tissues.