Automatic and observable spinal cord compression grading system for lung cancer spinal metastasis

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Epidural spinal cord compression (ESCC) assessment is crucial for lung cancer spinal metastasis surgery.
• Manual ESCC grading is subjective and inefficient, with complex MRI features and no public datasets.
• Developing an automated system for ESCC grading is needed to improve clinical assessment.

Purpose of the Study:

• To develop an automatic and efficient system for assessing spinal cord compression grading.
• To utilize multi-tissue and feature segmentation for improved grading accuracy.
• To address the lack of publicly labeled datasets for ESCC assessment.

Main Methods:

• Constructed a dataset of 577 lung cancer spinal metastasis MRI slices with segmentation masks and ESCC grades.
• Employed an encoder-decoder segmentation network with a pyramid pooling module and cross-attention for tissue segmentation.
• Utilized a pre-trained dual attention network (DAN) with transfer learning for accurate compression grading.

Main Results:

• Achieved average pixel accuracy of 0.819 and intersection over union of 0.711 for segmentation.
• Reached a classification accuracy of 0.95 and precision of 0.89 for ESCC grading.
• Demonstrated superior performance compared to traditional segmentation and classification models.

Conclusions:

• Developed an automated system for spinal cord compression grading in lung cancer spinal metastasis with 0.95 accuracy.
• The system enhances clinical assessment accuracy and efficiency for orthopedic surgeons.
• Provides a valuable tool for determining surgical intervention and treatment plans for patients.