Special Poster Session 51st International Society for the Study of the Lumbar Spine Annual Meeting 2025

Deep Learning-Based Object Detection Algorithm Using Magnetic Resonance Imaging for Differentiating Pathological Vertebral Fractures Caused by Malignant Metastatic Cancer in Patients with Vertebral Compression Fractures (116198)

Joonghyun Ahn 1 , Young Hoon Kim 2 , Jun-Seok Lee 3 , Kee Y Ha 2 , Hyung-youl Park 3 , June Lee 4 , Youjin Shin 4 , Jae Chul Lee 5
  1. Orthopedic Surgery, Bucheon St. Mary's Hospital, College of Medicine, The Catholic University of Korea, Bucheon, Gyunggi-do, Repulibc of Korea
  2. Orthopedic Surgery, Seoul St. Mary's Hospital, College of Medicine, The Catholic university of Korea, Seoul, Republic of Korea
  3. Orthopedic Surgery, Eunpyeong St. Mary's Hospital, College of Medicine, The Catholic University of Korea, Seoul, Republic of Korea
  4. Data Science, The Catholic University of Korea, Bucheon, Gyunggi-do, 대한민국
  5. Orthopaedic Surgery, Soonchunhyang University Seoul Hospital, College of Medicine, Soonchunhyang University, Seoul, Republic of Korea

Introduction

This study aims to detect pathological vertebral fractures caused by malignant metastatic cancer early through the latest deep learning object detection model, thereby improving the diagnostic efficiency of medical professionals and contributing to the establishment of treatment plans.

 

Methods

A dataset was created using sagittal-plane MRI scans of the thoracolumbar spine from a single institution. The data used in this study included 747 patients diagnosed with vertebral compression fractures (VCF) and 62 patients diagnosed with fractures caused by malignant metastatic cancer. The vertebrae with fractures were annotated by two spine specialists using the labelImg program. Then, the train, validation, and test datasets were divided in an 8:1:1 ratio, and the deep learning-based object detection model YOLOv8 was trained. The classification performance of the model was evaluated using Precision, Recall, and F1-score, and the object detection performance was assessed using mean Average Precision (mAP).

 

Results

The YOLOv8 model achieved precision (0.929), recall (0.928), F1-score (0.93) in differentiating vertebral fractures caused by benign origin and malignant metastatic cancer. The model's object detection ability demonstrated a performance with a mean average precision (mAP) of 0.967.

 

Discussion

The deep learning-based spinal disease diagnosis model developed in this study has shown high accuracy and efficiency, and which will be useful as a tool to help early screening for the possibility of metastatic cancer after MRI examination.