Deep-learning Based Classification of Spine CT

• Conditions: Surgical Procedure, Unspecified
• Interventions: Diagnostic Test: deep learning

• Registration Number: NCT03790930
• Lead Sponsor: Shanghai 10th People's Hospital

Brief Summary

It is time-consuming for spine surgeons or radiologists to conduct manual classifications of spinal CT, which may also be correlated with high inter-observer variance. With the development of computer science, deep learning has emerged as a promising technique to classify images from individual level to pixel level. The main of the study is to automatically identify and classify the lesions, or segment targeted structures on spinal CT with deep learning.

Detailed Description

Computer tomography (CT) is one of the most important imaging tool to assist the diagnostic and treatment of spinal disease. Classification of specific targets (e.g. individuals, lesions, etc.) is one of the most common mission of medical image analysis. However, it is time-consuming for spine surgeons or radiologists to conduct manual classifications of spinal CT, which may also be correlated with high inter-observer variance. With the development of computer science, deep learning has emerged as a promising technique to classify images from individual level to pixel level. The main of the study is to automatically identify and classify the lesions, or segment targeted structures on spinal CT with deep learning.

Study Timeline

• Study First Submitted: 2018-11-16
• Study First Submitted QC: 2018-12-29
• Study First Posted: 2019-01-02
• Study Start: 2019-02-22
• Status Verified: 2020-05
• Primary Completion: 2020-05
• Study Completion: 2020-05
• Last Update Submitted: 2020-05-10
• Last Update Posted: 2020-05-12

Recruitment & Eligibility

• Status: UNKNOWN
• Sex: All
• Target Recruitment: 500

Inclusion Criteria

• spinal thin layer CT

Exclusion Critera:

• medals or other implants induce artifact
• poor image quality

Exclusion Criteria

Not provided

Study & Design

• Study Type: OBSERVATIONAL
• Study Design: Not specified

Arm && Interventions

Group	Intervention	Description
thin layer CT	deep learning	Thin-layer CT will be manually labeled and used to train, validate and test deep learning algorithm.

Primary Outcome Measures

Name	Time	Method
segmentation accuracy	1 day	segmentation accuracy of multiple structures (e.g. Dice score, etc.)
classification accuracy	1 day	classification accuracy (e.g. area under the curve, etc.)

Trial Locations

Locations (1)

Shanghai Tenth People's Hospital; 🇨🇳 Shanghai, Shanghai, China