Deep Learning of Anterior Talofibular Ligament: Comparison of Different Models

• Conditions: Lateral Ligament, Ankle
• Interventions: Diagnostic Test: Diagnositic test

• Registration Number: NCT04955067
• Lead Sponsor: Peking University Third Hospital

Brief Summary

The purpose of this study is to study the injury of the anterior talofibular ligament by deep learning method and compare a variety of different deep learning models to establish a deep learning method that can accurately identify and grade the injury of anterior talofibular ligament, and obtain a model with better recognition and grading effect.

Detailed Description

1. Recognition and segmentation of anterior talofibular ligament based on DenseNet. Densenet was used to recognize the axial T2-fs image, and the image level was the most typical one. The labelimg program based on Python was used to locate the coordinates of the anterior talofibular ligament and then imported into Python for learning. All the data were divided into a training set (70%, and then 30% of the training set was selected as the verification set). The remaining 30% was used as the test set to evaluate the accuracy of model recognition. After identifying the anterior talofibular ligament, the local clipping and amplification are carried out to remove the redundant information. Finally, input the result to the next step.

2. Establishment and comparison of various deep learning models: four deep learning models were established and compared in this study, namely VGG19, AlexNet, CapsNet, and GoogleNet. The models using image fitting alone and those combining with clinical physical examination data were compared for each deep learning model. The diagnostic efficiency between models was expressed by the ROC curve, including AUC, F1 score, etc. the ROC curve was further analyzed by t-test, Delong test, and other statistical methods. In this study, the data were divided into a training set (70%, 30% in the training set as the validation set), and the remaining 30% as the test set to evaluate the classification accuracy.

Study Timeline

• Study Start: 2021-01-01
• Status Verified: 2021-06
• Study First Submitted: 2021-06-28
• Study First Submitted QC: 2021-06-29
• Last Update Submitted: 2021-06-29
• Study First Posted: 2021-07-08
• Last Update Posted: 2021-07-08
• Primary Completion: 2021-12-30
• Study Completion: 2022-03-30

Recruitment & Eligibility

• Status: UNKNOWN
• Sex: All
• Target Recruitment: 1000

Inclusion Criteria

1. Without any treatment before imaging examination;
2. MR of ankle joint was performed within 3 months before operation and the image quality was good;
3. Arthroscopic operation was performed in our hospital and the operation records were complete.

Read More

Exclusion Criteria

1. history of ankle surgery, history of cancer or previous fractures.
2. Unclear image, serious artifact or incomplete clinical data.

Read More

Study & Design

• Study Type: OBSERVATIONAL
• Study Design: Not specified

Arm && Interventions

Group	Intervention	Description
Ligament tear-Grade 2	Diagnositic test	Arthroscopy of the ankle joint revealed partial or complete loss of ligaments.
Normal control group-Grade 0	Diagnositic test	Arthroscopic examination of the ankle joint was normal, and the ligament was intact without injury or tear.
Ligament injury -Grade 1	Diagnositic test	Arthroscopic examination of the ankle joint showed ligament degeneration or injury, but no local or complete tear.

Primary Outcome Measures

Name	Time	Method
Deep Learning of Anterior Talofibular Ligament: Comparison of Different Models	2021.1-2022.3.1	The model of deep learning was obtained for diagnosis and grading of anterior fibular ligament and compared with the doctors of different grades.

Trial Locations

Locations (1)

Peking University Third Hospital; 🇨🇳 Beijing, Beijing, China