Multimodal Imaging-assisted Diagnosis Model for Cervical Spine Tumors

Spine Tumor Clinical Trial

Official title:

Based on a Small Sample Deep Learning Multi-modal Image-assisted Diagnosis Model of Cervical Spine Tumors Clinical Application Research

Clinical Trial Details — Status: Completed

Administrative data

• NCT number: NCT04959656
• Other study ID #: IRB00006761-M2020255
• Status: Completed
• Start date: January 1, 2020
• Est. completion date: June 1, 2021

Study information

• Verified date: July 2021
• Source: Peking University Third Hospital
• Contact: n/a
• Is FDA regulated: No
• Study type: Observational

Clinical Trial Summary

Cervical spine tumor is a small sample of tumor disease with low incidence, great harm, and complex anatomical structure. It is very difficult to identify and classify benign and malignant cervical spine tumors clinically. The deep learning model we constructed in the early stage has a higher accuracy rate for the image diagnosis of cervical spondylosis with a large number of cases, and a better clinical application effect, but the accuracy rate for cervical spine tumors with a small number of cases is lower. The reason may be the amount of data. With limited tasks, the traditional deep learning model is difficult to play an effective role. Based on this, we propose to build a small sample-oriented deep learning model to assist clinicians in the diagnosis of cervical spine tumors with multimodal images, and to evaluate the benign and malignant tumors.

Description:

Cervical spine tumor is a small-sample tumor disease with low incidence, great harm, and complex anatomical structure. It is very difficult to identify and classify benign and malignant cervical spine tumors clinically. The deep learning model we constructed in the early stage is suitable for the large number of cases. The imaging diagnosis of cervical spondylosis has a high accuracy rate and a good clinical application effect, but the accuracy rate is low for cervical spine tumors with a small number of cases. The reason may be that for tasks with limited amount of data, the traditional deep learning model is difficult to play an effective role. Based on this, we propose to construct a small sample-oriented deep learning model to assist clinicians in the diagnosis of cervical spine tumors in multi-modal imaging, and to evaluate the benign and malignant tumors. This research will not only improve the efficiency and efficiency of cervical spine tumor imaging diagnosis. Accuracy, to guide clinical personalized treatment, will also provide a basis for the clinical application of deep learning in the field of small samples, which has important clinical significance.

Recruitment information / eligibility

• Status: Completed
• Enrollment: 600
• Est. completion date: June 1, 2021
• Est. primary completion date: June 1, 2020
• Accepts healthy volunteers: No
• Gender: All
• Age group: 18 Years to 50 Years

Eligibility

Inclusion Criteria:

• 18-50 years old, about 300 males and females; in the orthopedics outpatient and emergency department of our hospital, the imaging scans (X-ray, CT, MR) showed no obvious abnormalities.

Exclusion Criteria:

• have had surgery before acquiring the images, Those who have cervical spine fractures, deformities, infections, etc. who cannot cooperate with imaging examinations, and those who have not signed the informed consent. The normal control group" includes about 600 patients with normal or slightly degenerated cervical spine, as a standard for training computers to recognize cervical spine structures Images and control images for detecting tumor lesions.

Related Conditions & MeSH terms

• Spine Tumor

Locations

Country	Name	City	State
China	Peking University Third Hospital	Beijing

Sponsors (1)

Lead Sponsor	Collaborator
Peking University Third Hospital

Country where clinical trial is conducted

China, 

Outcome

Type	Measure	Description	Time frame	Safety issue
Primary	tumor detection	On the basis of the cervical spine structure, it is the modeling of the tumor. The model based on weakly supervised learning recognizes the morphological features such as the size of the tumor lesion, and uses the fast-adapted meta-learning method to achieve a fast model under a small amount of training. Optimize, and finally evaluate the benignity, borderline and malignant probability of the tumor and use it as an output.	2022-2023
Secondary	cervical spine detection	Taking the postoperative pathology report of cancer patients as the audit standard, testing the sensitivity and accuracy of the model, and integrating it into a complete deep learning model.	2022-2023