In this retrospective, three-center, multireader study using consecutive knee radiographs in patients with suspected osteoarthritis, concurrent artificial intelligence assistance improved osteoarthritis grading performance of junior readers and increased interobserver agreement across all readers.
14 November 2013 in Scandinavian Journal of Medicine & Science in Sports
The first patient was misclassified in the diagnostic conclusion according to a local clinical expert opinion in a new clinical implementation of a knee osteoarthritis artificial intelligence (AI) algorithm at Bispebjerg-Frederiksberg University Hospital, Copenhagen, Denmark. In preparation for the evaluation of the AI algorithm, the implementation team collaborated with internal and external partners to plan workflows, and the algorithm was externally validated. After the misclassification, the team was left wondering: what is an acceptable error rate for a low-risk AI diagnostic algorithm? A survey among employees at the Department of Radiology showed significantly lower acceptable error rates for AI (6.8 %) than humans (11.3 %). A general mistrust of AI could cause the discrepancy in acceptable errors. AI may have the disadvantage of limited social capital and likeability compared to human co-workers, and therefore, less potential for forgiveness. Future AI development and implementation require further investigation of the fear of AI’s unknown errors to enhance the trustworthiness of perceiving AI as a co-worker. Benchmark tools, transparency, and explainability are also needed to evaluate AI algorithms in clinical implementations to ensure acceptable performance.
To create a scalable and feasible retrospective consecutive knee osteoarthritis (OA) radiographic database with limited human labor using commercial and custom-built artificial intelligence (AI) tools. We applied four AI tools, two commercially available and two custom-built tools, to analyze 6 years of clinical consecutive knee radiographs from patients aged 35-79 at the University of Copenhagen Hospital, Bispebjerg-Frederiksberg Hospital, Denmark. The tools provided Kellgren-Lawrence (KL) grades, joint space widths, patella osteophyte detection, radiographic view detection, knee joint implant detection, and radiographic marker detection. In total, 25,778 knee radiographs from 8575 patients were included in the database after excluding inapplicable radiographs, and 92.5% of the knees had a complete OA dataset. Using the four AI tools, we saved about 800 hours of radiologist reading time and only manually reviewed 16.0% of the images in the database. This study shows that clinical knee OA databases can be built using AI with limited human reading time for uniform grading and measurements. The concept is scalable temporally and across geographic regions and could help diversify further OA research by efficiently including radiographic knee OA data from different populations globally. We can prevent data dredging and overfitting OA theories on existing trite cohorts by including various gene pools and continuous expansion of new clinical cohorts. Furthermore, the suggested tools and applied approaches provide an ability to retest previous hypotheses and test new hypotheses on real-life clinical data with current disease prevalence and trends.
Humans have been shown to have biases when reading medical images, raising questions about whether humans are uniform in their disease gradings. Artificial intelligence (AI) tools trained on human-labeled data may have inherent human non-uniformity. In this study, we used a radiographic knee osteoarthritis external validation dataset of 50 patients and a six-year retrospective consecutive clinical cohort of 8,273 patients. An FDA-approved and CE-marked AI tool was tested for potential non-uniformity in Kellgren-Lawrence grades between the right and left sides of the images. We flipped the images horizontally so that a left knee looked like a right knee and vice versa. According to human review, the AI tool showed non-uniformity with 20–22% disagreements on the external validation dataset and 13.6% on the cohort. However, we found no evidence of a significant difference in the accuracy compared to senior radiologists on the external validation dataset, or age bias or sex bias on the cohort. AI non-uniformity can boost the evaluated performance against humans, but image areas with inferior performance should be investigated.
Background: Radiographic evaluation of knee osteoarthritis (KOA) commonly supports clinical findings. Ground truth is difficult to establish and concerns exist on the inter-and intrarater agreement of the findings. RBknee™ is a CE-marked and FDA-cleared AI tool for automatic assessment and reporting of radiographic KOA on standard projection radiographs. Objectives: To investigate how the use of an AI tool affects the accuracy among human readers across three European hospitals in grading the severity of osteoarthritis and associated individual radiographic features. In addition, the performance of the AI tool will also be compared to reference standards established by experts in a stand-alone validation. Methods: In this retrospective multicenter, fully-crossed, multi-reader, multi-case (MRMC) study, the AI support tool RBknee is introduced as a diagnostic intervention. Four Index Readers from each site (two orthopaedic surgeons and two radiologists) will read all studies twice in two runs separated by a washout period of at least four weeks. In both runs, the experiment will be arranged so that the AI-aid will be available for half of the images in the first session and for the second half of the images in the second session. The order of the images will be randomised in order to minimise temporal effects and biases. The primary endpoint is the difference in diagnostic test accuracy for radiographic KOA grading without and with the aid of the AI tool and will be measured as the ordinal weighted accuracy. Data: The data includes radiographic images from 225 studies (unique patients, retrospective data) with weight-bearing bilateral PA/AP and LAT projections of the symptomatic knee(s). Each site contributes to the cohort with 75 studies of which 70 will be consecutive and 5 will be selected to balance the prevalence of radiographic KOA severity. Reference standard: The reference standard will be established based on independent grading by three KOA Reference Experts and adjudicated by majority vote. Where impossible to resolve by majority voting, adjudication will be established by consensus. Index test, AI tool (stand-alone validation): The diagnostic accuracy of RBknee will be tested against the reference standard. Index test, Index Readers: The 12 readers will grade KL on the PA/AP projection and patellar osteophytes on the lateral projection. Administrative information: Title: The trial is titled “AutoRayValid-RBknee”. Protocol version: Revision History Funding: This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 954221 for the EIC SME Instrument project AutoRay. The work only reflects the authors’ view and the European Commission is not responsible for any use that may be made from the information it contains. Roles and responsibilities: Authors’ contributions: MWB, MB, EHGO, JV, and KGH initiated the study design and AL, JUN, KZ and HR helped with implementation. MWB, LE and MJL provided statistical expertise in clinical study design and MWB is conducting the primary statistical analysis. MWB, MJL and LE prepared the manuscript and all authors contributed to its refinement and approved the final manuscript. Sponsor Contact information: Trial Sponsor: Radiobotics ApS Contact name: Liv Egnell Address: Esplanaden 8C, 1263 Copenhagen K, Denmark Email: liv@radiobotics.com