Retuve: Automated Multi-Modality Analysis of Hip Dysplasia with Open Source AI Journal: Software Impacts

Adam McArthur, Stephanie Wichuk, Stephen Burnside, Andrew Kirby, Alexander Scammon, Damian Sol, Abhilash Hareendranathan, Jacob L. Jaremko · 2025

Read Retuve: Automated Multi-Modality Analysis of Hip Dysplasia with Open Source AI Journal: Software Impacts by Adam McArthur, Stephanie Wichuk, Stephen Burnside, Andrew Kirby, Alexander Scammon, Damian Sol, Abhilash Hareendranathan, Jacob L. Jaremko (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-06422.txt.

Introduction

Eight authors, led by Adam McArthur, present Retuve in this 2025 paper, an open-source AI framework for hip dysplasia analysis. Developmental dysplasia of the hip is a birth condition in which the hip joint forms poorly. Finding it early matters, yet screening lacks a standard method.

AI studies of the condition have suffered from scarce data and code that others cannot rerun. Retuve answers with a full, open workflow. It gives out expert-marked ultrasound and X-ray image sets. It also gives pre-trained models with their training code and weights, plus a friendly Python interface. Its models segment the joint and find key landmarks. From those, the system measures standard diagnostic values, such as the alpha angle and the acetabular index. The aim is screening that any clinic can check and reuse.

Catalog record

CREATORAdam McArthur, Stephanie Wichuk, Stephen Burnside, Andrew Kirby, Alexander Scammon, Damian Sol, Abhilash Hareendranathan, Jacob L. Jaremko
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE25 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-06422

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Retuve: Automated Multi-Modality Analysis of Hip Dysplasia with Open Source AI Journal: Software Impacts Adam McArthur, Stephanie Wichuk, Stephen Burnside, Andrew Kirby, Alexander Scammon, Damian Sol, Abhilash Hareendranathan, Jacob L. Jaremko Abstract Developmental dysplasia of the hip ( DDH ) poses significant diagnostic challenges, hindering timely intervention. Current screening methodologies lack standardization, and AI-driven studies suffer from reproducibility issues due to limited data and code availability. To address these limitations, we introduce Retuve, an open-source framework for multi-modality DDH analysis, encompassing both ultrasound ( US ) and X-ray imaging. Retuve provides a complete and reproducible workflow, offering open datasets comprising expert-annotated US and X-ray images, pre-trained models with training code and weights, and a user-friendly Python Application Programming Interface ( API ). The framework integrates segmentation and landmark detection models, enabling automated measurement of key diagnostic parameters such as the alpha angle and acetabular index. By adhering to open-source principles, Retuve promotes transparency, collaboration, and accessibility in DDH research. This initiative has the potential to democratize DDH screening, facilitate early diagnosis, and ultimately improve patient outcomes by enabling widespread screening and early intervention. The GitHub repository/code can be found here: https://github.com/radoss-org/retuve † † graphicalabstract: † † highlights: Creation of an open-source framework facilitating ongoing research in DDH imaging, promoting collaborative advancement in the field. Development and release of a pioneering DDH open-source dataset, complete with expert annotations for both ultrasound and X-ray imaging modalities. Implementation of a modular AI system that seamlessly integrates segmentation and landmark models into Retuve, with three published plugins demonstrating its versatility. Nomenclature Acronym Definition AI Artificial Intelligence DDH Developmental Dysplasia of the Hip FP False Positive GHDR Global Hip Dysplasia Registry ICML International Conference on Machine Learning MICCAI Medical Image Computing and Computer Assisted Intervention OSI Open Source Initiative PACS Picture Archiving and Communication System POCUS Point-of-Care Ultrasound RCT Randomized Controlled Trial US Ultrasound VicHip Victorian Hip Dysplasia Registry MIMIC-III The Medical Information Mart for Intensive Care III LLM Large Language Model API Application Programming Interface MRI Magnetic Resonance Imaging CT Computed Tomography GUI Graphical User Interface CLI Command Line Interface CPU Central Processing Unit RAM Random Access Memory FPS Frames Per Second Table 1: List of acronyms used in this paper. 1 Introduction Developmental dysplasia of the hip ( DDH ) research is significantly hampered by two key challenges: the absence of standardized, objective screening methodologies and limitations conducting robust, reproducible AI-driven studies. The definition threshold for DDH is unclear. There are high costs to universal ultrasonographic screening programs, varying reported incidence rates from 1.5 to 20 per 1000 births [ 1 ] , and inconsistent risk factor assessments, such as breech position carrying a 2-24 times increased risk of DDH [ 2 , 3 , 4 , 5 , 6 ] . Despite high sensitivity rates of 88-95% for well-performed ultrasound screening [ 7 ] , the condition faces challenges typical of rare diseases. Recent studies report that False Positive ( FP ) cases are very costly for screening and occur at a high rate [ 8 ] . Up to 97% of ultrasonographically detected immature hips normalize naturally without intervention [ 9 , 10 ] , while global screening protocols lack consensus on issues such as who to screen, when and how [ 11 ] . The need for long-term randomized controlled trials ( RCTs ) extending into later life has been emphasized to validate screening and treatment protocols [ 11 ] . There are multiple large-scale efforts to try and gather the data required for these RCT s, such as the Global Hip Dysplasia Registry ( GHDR ) [ 12 ] and the Victorian Hip Dysplasia Registry ( VicHip ) [ 13 ] . Open data is vital for advancing reproducibility, transparency, and innovation in open-source AI research, particularly in medicine, where reliability and generalizability are critical. The reproducibility crisis in AI, driven by limited access to datasets, code, and methodologies, highlights the need for open data. Studies show that less than a third of AI research is reproducible, with inaccessible healthcare datasets being a major barrier [ 14 , 15 , 16 ] . For example, only 26% of AI studies in a systematic review were computationally reproducible, and this figure rose to 86% when both code and data were shared [ 15 ] . Similarly, in medical AI research, up to 97% of studies lacked sufficient transparency for real-world application [ 17 ] . Open data initiatives like MIMIC-III and OpenML have improved reproducibility by enabling validation across diverse populations and settings. MIMIC-III provides de-identified health records for over 40,000 patients, fostering collaboration and reproducibility in clinical research [ 18 , 19 ] . OpenML facilitates the sharing of machine learning datasets and experiments, promoting transparent benchmarking and algorithm development [ 20 , 21 ] . Leading conferences like MICCAI and ICML emphasize open data’s value. MICCAI 2024 has launched an Open Data initiative focusing on underrepresented populations to address global healthcare challenges [ 22 , 23 ] . ICML promotes "data-centric AI," highlighting the importance of high-quality datasets for impactful research [ 24 ] . Openly sharing data and making it easily available improves the quality, reliability, and usefulness of AI research in medicine. This approach not only mitigates biases but also accelerates innovation by enabling researchers worldwide to validate and build upon existing work [ 25 , 26 , 27 ] . Thus, we present Retuve, an open-source framework for DDH analysis of both ultrasound and X-ray imaging that adheres to the OSI definition of open-source AI [ 28 ] . Retuve provides complete reproducibility of results, offering open datasets and models (training code and weights), and a Python Application Programming Interface ( API ). This commitment to transparency and reproducibility empowers researchers and clinicians to collaboratively advance the field of automated DDH detection. Retuve’s white-box algorithm allows for detailed analysis and improvement of AI models for ultrasound and X-ray Screening. Retuve will continue to expand its open-source data and AI literature coverage over the next 2-3 years, fostering collaborative efforts to advance our understanding and treatment of DDH . Retuve is in line with similar efforts in the LLM community to provide fully-open source AI [ 29 ] . Retuve aims to support widespread screening and early intervention for DDH . This can potentially reduce the reliance on highly skilled professionals, facilitate earlier diagnosis, and ultimately prevent early-onset osteoarthritis linked to DDH . In the future, we hope to combine Retuve and Clinical Outcomes such as is now being acquired at GDHR and VicHip to create comprehensive AI models that can detect and risk-stratify DDH in a diverse range of settings, this is shown in Figure 1 . This includes automatic classification of Graf class (normal to severely dysplastic) and scan quality. Figure 1: Retuve can analyse ultrasounds and X-rays to generate conventional and new indices quantifying hip anatomy, such as acetabular alpha angle and coverage (ultrasound), and acetabular index (X-ray). These indices can be used to train the next generation of AI Models to diagnose DDH . 2 Code Metadata Item Description Current code version v0.1.0 Permanent link to code/repository https://github.com/radoss-org/retuve Latest commit at time of Publication fd4d10fd1bfa9ee8dd148e0af27dabbec2e12840 Legal Code License Apache V2.0 Licence Code

Read the complete text