Can an artificial intelligence powered software reliably assess pelvic radiographs?

DXA images vs. pelvic radiographs: Reliability of hip morphology measurements

OBJECTIVE

Dual-energy x-ray absorptiometry (DXA) images are increasingly used to study hip morphology. Whether hip morphology measurements are consistent between DXA images and radiographs is unknown. Therefore, we investigated the agreement and reliability of the measurements performed on DXA images and radiographs.

DESIGN

We included participants from the Rotterdam study, a population-based cohort study, who received a hip DXA image and pelvic radiograph on the same day. The acetabular depth-width ratio (ADR), modified acetabular index (mAI), alpha angle (AA), Wiberg and lateral center edge angle (WCEA, LCEA), extrusion index (EI) and triangular index ratio (TIR) were automatically determined on both imaging modalities. The intraobserver and intermethod agreement were studied using Bland-Altman methods, and the reliability was assessed using intraclass correlation coefficients (ICC). Secondly, the diagnostic agreement regarding dysplasia, cam, and pincer morphology was assessed using percent agreement and Cohen's kappa.

RESULTS

A total of 750 hips from 411 individuals, median age 67.3 years (range 52.2 - 90.6), 45.5% male, were included. The following intermethod ICCs (95% CI) were obtained: ADR 0.85 (0.74-0.91), mAI 0.75 (0.52-0.85), AA 0.72 (0.68-0.75), WCEA 0.81 (0.74-0.85), LCEA 0.93 (0.91-0.94), EI 0.88 (0.84-0.91), and TIR 0.81 (0.79-0.84). We found comparable intraobserver ICCs for each morphological measurement.

CONCLUSION

DXA images and pelvic radiographs could both reliably be used to study hip morphology. Due to the lower radiation burden, DXA images could be an excellent alternative to pelvic radiographs for research purposes.

Validation of AI-driven measurements for hip morphology assessment

RATIONALE AND OBJECTIVES

Accurate assessment of hip morphology is crucial for the diagnosis and management of hip pathologies. Traditional manual measurements are prone to mistakes and inter- and intra-reader variability. Artificial intelligence (AI) could mitigate such issues by providing accurate and reproducible measurements. The aim of this study was to compare the performance of BoneMetrics (Gleamer, Paris, France) in measuring pelvic and hip parameters on anteroposterior (AP) and false profile radiographs to expert manual measurements.

MATERIALS AND METHODS

This retrospective study included AP and false profile pelvic radiographs collected from private practices in France. Pelvic and hip measurements included the femoral neck shaft angle, lateral center edge angle, acetabular roof angle, pelvic obliquity, and vertical center anterior angle. AI measurements were compared to a ground truth established by two expert radiologists. Performance metrics included mean absolute error (MAE), Bland-Altman analysis, and intraclass correlation coefficients (ICC).

RESULTS

AI measurements were performed on AP views from 88 patients and on false profile views from 60 patients. They demonstrated high accuracy, with MAE values inferior to 0.5 mm for pelvic obliquity and inferior to 4.2° for all pelvic angles on both views. ICC values indicated good to excellent agreement between AI measurements and the ground truth (0.78-0.99). Notably, no significant differences were found in AI measurement accuracy between patients with normal and abnormal acetabular coverage.

CONCLUSION

The application of AI in measuring pelvic and hip parameters on AP and false profile radiographs demonstrates promising outcomes. The results reveal that these AI-powered measurements provide accurate estimations and show strong agreement with expert manual measurements. Ultimately, the use of AI has the potential to enhance the reproducibility of measurements as part of comprehensive hip assessments, thereby improving diagnostic accuracy.

Evaluation of a deep learning software for automated measurements on full-leg standing radiographs

BACKGROUND

Precise lower limb measurements are crucial for assessing musculoskeletal health; fully automated solutions have the potential to enhance standardization and reproducibility of these measurements. This study compared the measurements performed by BoneMetrics (Gleamer, Paris, France), a commercial artificial intelligence (AI)-based software, to expert manual measurements on anteroposterior full-leg standing radiographs.

METHODS

A retrospective analysis was conducted on a dataset comprising consecutive anteroposterior full-leg standing radiographs obtained from four imaging institutions. Key anatomical landmarks to define the hip-knee-ankle angle, pelvic obliquity, leg length, femoral length, and tibial length were annotated independently by two expert musculoskeletal radiologists and served as the ground truth. The performance of the AI was compared against these reference measurements using the mean absolute error, Bland-Altman analyses, and intraclass correlation coefficients.

RESULTS

A total of 175 anteroposterior full-leg standing radiographs from 167 patients were included in the final dataset (mean age = 49.9 ± 23.6 years old; 103 women and 64 men). Mean absolute error values were 0.30° (95% confidence interval [CI] [0.28, 0.32]) for the hip-knee-ankle angle, 0.75 mm (95% CI [0.60, 0.88]) for pelvic obliquity, 1.03 mm (95% CI [0.91,1.14]) for leg length from the top of the femoral head, 1.45 mm (95% CI [1.33, 1.60]) for leg length from the center of the femoral head, 0.95 mm (95% CI [0.85, 1.04]) for femoral length from the top of the femoral head, 1.23 mm (95% CI [1.12, 1.32]) for femoral length from the center of the femoral head, and 1.38 mm (95% CI [1.21, 1.52]) for tibial length. The Bland-Altman analyses revealed no systematic bias across all measurements. Additionally, the software exhibited excellent agreement with the gold-standard measurements with intraclass correlation coefficient (ICC) values above 0.97 for all parameters.

CONCLUSIONS

Automated measurements on anteroposterior full-leg standing radiographs offer a reliable alternative to manual assessments. The use of AI in musculoskeletal radiology has the potential to support physicians in their daily practice without compromising patient care standards.

Artificial intelligence in musculoskeletal imaging: realistic clinical applications in the next decade

This article will provide a perspective review of the most extensively investigated deep learning (DL) applications for musculoskeletal disease detection that have the best potential to translate into routine clinical practice over the next decade. Deep learning methods for detecting fractures, estimating pediatric bone age, calculating bone measurements such as lower extremity alignment and Cobb angle, and grading osteoarthritis on radiographs have been shown to have high diagnostic performance with many of these applications now commercially available for use in clinical practice. Many studies have also documented the feasibility of using DL methods for detecting joint pathology and characterizing bone tumors on magnetic resonance imaging (MRI). However, musculoskeletal disease detection on MRI is difficult as it requires multi-task, multi-class detection of complex abnormalities on multiple image slices with different tissue contrasts. The generalizability of DL methods for musculoskeletal disease detection on MRI is also challenging due to fluctuations in image quality caused by the wide variety of scanners and pulse sequences used in routine MRI protocols. The diagnostic performance of current DL methods for musculoskeletal disease detection must be further evaluated in well-designed prospective studies using large image datasets acquired at different institutions with different imaging parameters and imaging hardware before they can be fully implemented in clinical practice. Future studies must also investigate the true clinical benefits of current DL methods and determine whether they could enhance quality, reduce error rates, improve workflow, and decrease radiologist fatigue and burnout with all of this weighed against the costs.

Reliability and agreement of manual and automated morphological radiographic hip measurements

Objective

To determine the reliability and agreement of manual and automated morphological measurements, and agreement in morphological diagnoses.

Methods

Thirty pelvic radiographs were randomly selected from the World COACH consortium. Manual and automated measurements of acetabular depth-width ratio (ADR), modified acetabular index (mAI), alpha angle (AA), Wiberg center edge angle (WCEA), lateral center edge angle (LCEA), extrusion index (EI), neck-shaft angle (NSA), and triangular index ratio (TIR) were performed. Bland-Altman plots and intraclass correlation coefficients (ICCs) were used to test reliability. Agreement in diagnosing acetabular dysplasia, pincer and cam morphology by manual and automated measurements was assessed using percentage agreement. Visualizations of all measurements were scored by a radiologist.

Results

The Bland-Altman plots showed no to small mean differences between automated and manual measurements for all measurements except for ADR. Intraobserver ICCs of manual measurements ranged from 0.26 (95%-CI 0-0.57) for TIR to 0.95 (95%-CI 0.87-0.98) for LCEA. Interobserver ICCs of manual measurements ranged from 0.43 (95%-CI 0.10-0.68) for AA to 0.95 (95%-CI 0.86-0.98) for LCEA. Intermethod ICCs ranged from 0.46 (95%-CI 0.12-0.70) for AA to 0.89 (95%-CI 0.78-0.94) for LCEA. Radiographic diagnostic agreement ranged from 47% to 100% for the manual observers and 63%-96% for the automated method as assessed by the radiologist.

Conclusion

The automated algorithm performed equally well compared to manual measurement by trained observers, attesting to its reliability and efficiency in rapidly computing morphological measurements. This validated method can aid clinical practice and accelerate hip osteoarthritis research.

Automated radiographic hip morphology measurements: An open-access method

Objective

The aim of this study is to present a newly developed automated method to determine radiographic measurements of hip morphology on dual-energy x-ray absorptiometry (DXA) images. The secondary aim was to compare the performance of the automated and manual measurements.

Design

30 DXA scans from 13-year-olds of the prospective population-based cohort study Generation R were randomly selected. The hip shape was outlined automatically using radiographic landmarks from which the acetabular depth-width ratio (ADR), acetabular index (AI), alpha angle (AA), Wiberg and lateral center edge angle (WCEA) (LCEA), extrusion index (EI), neck-shaft angle (NSA), and the triangular index (TI) were determined. Manual assessments were performed twice by two orthopedic surgeons. The agreement within and between observers and methods was visualized using Bland-Altman plots, and the reliability was studied using the intraclass correlation coefficient (ICC) with 95 % confidence intervals (CI).

Results

The automated method was able to perform all radiographic hip morphology measurements. The intermethod reliability between the automated and manual measurements ranged from 0.57 to 0.96 and was comparable to or better than the manual interobserver reliability, except for the AI.

Conclusion

This open-access, automated method allows fast and reproducible calculation of radiographic measurements of hip morphology on right hip DXA images. It is a promising tool for performing automated radiographic measurements of hip morphology in large population studies and clinical practice.

Development of a diagnostic support system for distal humerus fracture using artificial intelligence

PURPOSE

AI has shown promise in automating and improving various tasks, including medical image analysis. Distal humerus fractures are a critical clinical concern that requires early diagnosis and treatment to avoid complications. The standard diagnostic method involves X-ray imaging, but subtle fractures can be missed, leading to delayed or incorrect diagnoses. Deep learning, a subset of artificial intelligence, has demonstrated the ability to automate medical image analysis tasks, potentially improving fracture identification accuracy and reducing the need for additional and cost-intensive imaging modalities (Schwarz et al. 2023). This study aims to develop a deep learning-based diagnostic support system for distal humerus fractures using conventional X-ray images. The primary objective of this study is to determine whether deep learning can provide reliable image-based fracture detection recommendations for distal humerus fractures.

METHODS

Between March 2017 and March 2022, our tertiary hospital's PACS data were evaluated for conventional radiography images of the anteroposterior (AP) and lateral elbow for suspected traumatic distal humerus fractures. The data set consisted of 4931 images of patients seven years and older, after excluding paediatric images below seven years due to the absence of ossification centres. Two senior orthopaedic surgeons with 12 + years of experience reviewed and labelled the images as fractured or normal. The data set was split into training sets (79.88%) and validation tests (20.1%). Image pre-processing was performed by cropping the images to 224 × 224 pixels around the capitellum, and the deep learning algorithm architecture used was ResNet18.

RESULTS

The deep learning model demonstrated an accuracy of 69.14% in the validation test set, with a specificity of 95.89% and a positive predictive value (PPV) of 99.47%. However, the sensitivity was 61.49%, indicating that the model had a relatively high false negative rate. ROC analysis showed an AUC of 0.787 when deep learning AI was the reference and an AUC of 0.580 when the most senior orthopaedic surgeon was the reference. The performance of the model was compared with that of other orthopaedic surgeons of varying experience levels, showing varying levels of diagnostic precision.

CONCLUSION

The developed deep learning-based diagnostic support system shows potential for accurately diagnosing distal humerus fractures using AP and lateral elbow radiographs. The model's specificity and PPV indicate its ability to mark out occult lesions and has a high false positive rate. Further research and validation are necessary to improve the sensitivity and diagnostic accuracy of the model for practical clinical implementation.

Artificial intelligence-based three-dimensional templating for total joint arthroplasty planning: a scoping review

PURPOSE

The purpose of this review is to evaluate the current status of research on the application of artificial intelligence (AI)-based three-dimensional (3D) templating in preoperative planning of total joint arthroplasty.

METHODS

This scoping review followed the PRISMA, PRISMA-ScR guidelines, and five stage methodological framework for scoping reviews. Studies of patients undergoing primary or revision joint arthroplasty surgery that utilised AI-based 3D templating for surgical planning were included. Outcome measures included dataset and model development characteristics, AI performance metrics, and time performance. After AI-based 3D planning, the accuracy of component size and placement estimation and postoperative outcome data were collected.

RESULTS

Nine studies satisfied inclusion criteria including a focus on computed tomography (CT) or magnetic resonance imaging (MRI)-based AI templating for use in hip or knee arthroplasty. AI-based 3D templating systems reduced surgical planning time and improved implant size/position and imaging feature estimation compared to conventional radiographic templating. Several components of data processing and model development and testing were insufficiently covered in the studies included in this scoping review.

CONCLUSIONS

AI-based 3D templating systems have the potential to improve preoperative planning for joint arthroplasty surgery. This technology offers more accurate and personalized preoperative planning, which has potential to improve functional outcomes for patients. However, deficiencies in several key areas, including data handling, model development, and testing, can potentially hinder the reproducibility and reliability of the methods proposed. As such, further research is needed to definitively evaluate the efficacy and feasibility of these systems.

Fully automated assessment of the knee alignment on long leg radiographs following corrective knee osteotomies in patients with valgus or varus deformities

INTRODUCTION

The assessment of the knee alignment on long leg radiographs (LLR) postoperative to corrective knee osteotomies (CKOs) is highly dependent on the reader's expertise. Artificial Intelligence (AI) algorithms may help automate and standardise this process. The study aimed to analyse the reliability of an AI-algorithm for the evaluation of LLRs following CKOs.

MATERIALS AND METHODS

In this study, we analysed a validation cohort of 110 postoperative LLRs from 102 patients. All patients underwent CKO, including distal femoral (DFO), high tibial (HTO) and bilevel osteotomies. The agreement between manual measurements and the AI-algorithm was assessed for the mechanical axis deviation (MAD), hip knee ankle angle (HKA), anatomical-mechanical-axis-angle (AMA), joint line convergence angle (JLCA), mechanical lateral proximal femur angle (mLPFA), mechanical lateral distal femoral angle (mLDFA), mechanical medial proximal tibia angle (mMPTA) and mechanical lateral distal tibia angle (mLDTA), using the intra-class-correlation (ICC) coefficient between the readers, each reader and the AI and the mean of the manual reads and the AI-algorithm and Bland-Altman Plots between the manual reads and the AI software for the MAD, HKA, mLDFA and mMPTA.

RESULTS

In the validation cohort, the AI software showed excellent agreement with the manual reads (ICC: 0.81-0.99). The agreement between the readers (Inter-rater) showed excellent correlations (ICC: 0.95-0. The mean difference in the DFO group for the MAD, HKA, mLDFA and mMPTA were 0.50 mm, - 0.12°, 0.55° and 0.15°. In the HTO group the mean difference for the MAD, HKA, mLDFA and mMPTA were 0.36 mm, - 0.17°, 0.57° and 0.08°, respectively. Reliable outputs were generated in 95.4% of the validation cohort.

CONCLUSION

 he application of AI-algorithms for the assessment of lower limb alignment on LLRs following CKOs shows reliable and accurate results.

LEVEL OF EVIDENCE

Diagnostic Level III.

Predicting Outcomes in Hip Arthroscopy for Femoroacetabular Impingement Syndrome

PURPOSE OF REVIEW

Arthroscopic treatment of femoroacetabular impingement syndrome (FAIS) continues to rise in incidence, and thus there is an increased focus on factors that predict patient outcomes. The factors that impact the outcomes of arthroscopic FAIS treatment are complex. The purpose of this review is to outline the current literature concerning predictors of patient outcomes for arthroscopic treatment of FAIS.

RECENT FINDINGS

Multiple studies have shown that various patient demographics, joint parameters, and surgical techniques are all correlated with postoperative outcomes after arthroscopic FAIS surgery, as measured by both validated patient-reported outcome (PRO) scores and rates of revision surgery including hip arthroplasty. To accurately predict patient outcomes for arthroscopic FAIS surgery, consideration should be directed toward preoperative patient-specific factors and intraoperative technical factors. The future of accurately selecting patient predictors for outcomes will only improve with increased data, improved techniques, and technological advancement.

Gender-specific distribution of knee morphology according to CPAK and functional phenotype classification: analysis of 8739 osteoarthritic knees prior to total knee arthroplasty using artificial intelligence

PURPOSE

Osteoarthritis of the knee is commonly associated with malalignment of the lower limb. Recent classifications, as the Coronal Plane Alignment of the Knee (CPAK) and Functional Phenotype classification, describe the bony knee morphology in addition to the overall limb alignment. Data on distribution of these classifications is not sufficient in large populations. The aim of this study was to analyse the preoperative knee morphology with regard to the aforementioned classifications in long leg radiographs prior to total knee arthroplasty surgery using Artificial Intelligence.

METHODS

The cohort comprised 8739 preoperative long leg radiographs of 7456 patients of all total knee arthroplasty surgeries between 2009 and 2021 from our institutional database. The automated measurements were performed with the validated Artificial Intelligence software LAMA (ImageBiopsy Lab, Vienna) and included standardized axes and angles [hip-knee-ankle angle (HKA), mechanical lateral distal femur angle (mLDFA), mechanical medial proximal tibia angle (mMPTA), mechanical axis deviation (MAD), anatomic mechanic axis deviation (AMA) and joint line convergence angle (JLCA)]. CPAK and functional phenotype classifications were performed and all measurements were analysed for gender, age, and body mass index (BMI) within these subgroups.

RESULTS

Varus alignment was more common in men (m: 2008, 68.5%; f: 2953, 50.8%) while neutral (m: 578, 19.7%; f: 1357, 23.4%) and valgus (m: 345, 11.8%; f: 1498, 25.8%) alignment was more common in women. The most common morphotypes according to CPAK classification were CPAK Type I (2454; 28.1%), Type II (2383; 27.3%), and Type III (1830; 20.9%). An apex proximal joint line (CPAK Type VII, VIII and IX) was only found in 1.3% of all cases (n = 121). In men, CPAK Type I (1136; 38.8%) and CPAK Type II (799; 27.3%) were the most common types and women were spread more equally between CPAK Type I (1318; 22.7%), Type II (1584; 27.3%) and Type III (1494; 25.7%) (p < 0.001). The most common combination of femur and tibia types was NEUmLDFA0°,NEUmMPTA0° (m: 514, 17.5%; f: 1004, 17.3%), but men showed femoral varus more often. Patients with a higher BMI showed a significantly lower age at surgery (R2 = 0.09, p < 0.001). There were significant differences between men and women for all radiographic parameters (p < 0.001).

CONCLUSION

Distribution in knee morphology with gender-specific differences highlights the wide range in osteoarthritic knees, characterized by CPAK and phenotype classification and may influence future surgical planning.

LEVEL OF EVIDENCE

Level III.