AI-Based Screening and Diagnostic Models for Femoroacetabular Impingement Syndrome (FAIS-AI)

Development and Validation of AI-Based Screening and Diagnostic Models for Femoroacetabular Impingement Syndrome: A Multicenter Study Integrating Clinical Features and Pelvic Radiographs

Status

Completed

Phases

Unknown

Study type

Observational

Source

ClinicalTrials.gov

Registry ID

NCT07737223

Acronym

FAIS-AI

Enrollment

2617

Registered

2026-07-30

Start date

2019-01-01

Completion date

2025-12-30

Last updated

2026-07-30

For informational purposes only — not medical advice. Sourced from public registries and may not reflect the latest updates. Terms

Conditions

Femoroacetabular Impingement Syndrome

Keywords

Femoroacetabular Impingement, FAIS, Artificial Intelligence, Deep Learning, Hip Pain, Screening Model, Diagnostic Model, Pelvic Radiograph, Neural Network, YOLOv8, Convolutional Neural Network, CenterNet, SHAP, Hip X-ray

Brief summary

Femoroacetabular impingement syndrome (FAIS) is the leading cause of hip pain in young adults and frequently progresses to osteoarthritis, often exacerbated by delayed diagnosis in primary care. Current AI models for FAIS diagnosis primarily rely on single imaging modalities, limiting their diagnostic accuracy and clinical utility. This multicenter, retrospective-prospective study aims to develop and validate AI-based screening and diagnostic models for FAIS by integrating multimodal clinical features and pelvic radiographic data. A retrospective cohort of 1,841 patients (January 2019 to January 2025) was collected from four tertiary centers in Beijing (First and Fourth Medical Centers of PLA General Hospital, Beijing Friendship Hospital, and Rocket Force Characteristic Medical Center) for model development and internal validation. A screening model was built using the 10 most contributory clinical features (identified via SHAP analysis from 47 consensus-based features) with a fully connected neural network. A diagnostic model was built by combining clinical features, automated hip radiographic measurements (CE Angle, Tonnis Angle, Alpha Angle, Femoral Neck-Shaft Angle via CenterNet), and hip X-ray images (via YOLOv8 + CNN) through a dual-channel hybrid deep learning architecture. Prospective external validation was performed on an independent cohort of 776 patients from four population groups (large hospital, athletic, student, community) between February and November 2025. Model performance was evaluated using AUC, sensitivity, specificity, accuracy, PPV, NPV, and decision curve analysis, and compared against five physicians of varying seniority. The study aims to address FAIS diagnostic delays by providing an AI-based solution suitable for patient self-assessment, primary care screening, and specialist referral decision-making.

Interventions

None listed

Study design

Observational model

COHORT

Time perspective

OTHER

Eligibility

Sex/Gender

ALL

Age

12 Years to No maximum

Healthy volunteers

Inclusion criteria

Patients presenting to the outpatient clinic with a chief complaint of hip pain Meeting preliminary clinical suspicion of hip pathology (based on history and physical examination) Willing and able to provide written informed consent

Exclusion criteria

Groin or thigh hematoma, or abdominal/pelvic masses (identified on physical examination or imaging) Non-musculoskeletal conditions causing hip-region pain (e.g., urinary tract disorders, gynecological conditions) Signs of active infection (fever with elevated C-reactive protein) Incomplete or substandard clinical or imaging data (e.g., poor-quality radiographs, missing key variables)

Design outcomes

Primary

Measure	Time frame	Description
Area Under the Receiver Operating Characteristic Curve (AUC) of the AI screening model for identifying FAIS	Through study completion, up to 7 years	The AI screening model integrates 10 key clinical features identified through SHAP analysis using a fully connected neural network. AUC will be calculated from the receiver operating characteristic (ROC) curve, with values ranging from 0.5 (no discrimination) to 1.0 (perfect discrimination), to evaluate the screening model diagnostic performance.
Area Under the Receiver Operating Characteristic Curve (AUC) of the AI diagnostic model for identifying FAIS	Through study completion, up to 7 years	The AI diagnostic model combines clinical features, automated hip radiographic measurements (CE Angle, Tonnis Angle, Alpha Angle, Femoral Neck-Shaft Angle via CenterNet), and hip X-ray images (via YOLOv8 + CNN) through a dual-channel hybrid deep learning architecture. AUC will be calculated from the ROC curve to evaluate the comprehensive diagnostic performance.

Secondary

Measure	Time frame	Description
Sensitivity, Specificity, Positive Predictive Value (PPV), and Negative Predictive Value (NPV) of the AI screening and diagnostic models	Through study completion, up to 7 years	Sensitivity, specificity, PPV, and NPV will be calculated at the optimal threshold determined from the ROC curve analysis (Youden index). These metrics provide clinically meaningful measures of the models diagnostic accuracy for FAIS detection in real-world clinical settings.
Net benefit of the AI models in Decision Curve Analysis (DCA)	Through study completion, up to 7 years	Decision curve analysis will be performed to assess the clinical net benefit of the AI screening and diagnostic models across a range of threshold probabilities. This analysis evaluates whether using the AI models for clinical decision-making provides greater benefit than treating all or treating none.
Comparison of AUC between the AI models and clinicians of varying seniority	Through study completion, up to 7 years	The AUC of the AI models will be compared against the diagnostic performance of five physicians with varying levels of clinical experience (ranging from junior resident to senior attending physician) using DeLong test. Statistical significance will be set at p less than 0.05.
Intraclass Correlation Coefficient (ICC) of automated hip radiographic measurements	Through study completion, up to 7 years	The agreement between automated measurements (CE Angle, Tonnis Angle, Alpha Angle, Femoral Neck-Shaft Angle obtained via CenterNet) and manual measurements performed by two independent radiologists will be evaluated using the Intraclass Correlation Coefficient (ICC). ICC values greater than 0.75 indicate good reliability and greater than 0.90 indicate excellent reliability.

Countries

China

Outcome results