Radiology is an important medical service, deploying medical imaging to diagnose diseases and guide treatment within the bodies of humans. The practice is grouped into diagnostic radiology and interventional radiology. AI is increasingly being researched as a potential adjunct to radiologist-led interpretation of imaging.
Training of deep learning models can use either supervised learning (presenting pairs of inputs and desired outputs), unsupervised learning (the system clusters the data in classes), or reinforcement learning (the system learns by being rewarded). Radiomics is an extension of these processes; this refers to the extraction of a large number of features from medical images using data-characterisation algorithms.
Some recent advances in medical artificial intelligence have demonstrated how AI enhances radiologist accuracy, speeds diagnoses, and improves patient outcomes.
SimonMed Imaging, a U.S. outpatient imaging provider, has presented two pieces of AI research on fracture detection at the Radiological Society of North America’s Radiology Conference and Annual Meeting 2024.
The two datasets highlight how AI improves bone fracture detection in geriatric and paediatric patients by reducing turnaround times and enhancing radiologist performance.
The first study evaluated the effectiveness of implementing an enterprise wide artificial intelligence (AI) system for fracture detection from radiographs The goal was to assess the impact of AI-powered worklist prioritization on reducing turnaround time.
For this exercise, 26,690 MSK (musculoskeletal) radiographs detecting fractures were assessed, ultimately finding that 10.6 percent were detected during the pre-AI period and 14.7 percent during the post-AI period, ultimately shortening turnaround time while improving quality.
This study demonstrated how AI-powered tools enhance detection accuracy and can assist in streamlining workflows, leading to faster and more reliable patient care.
The second study focuses on paediatric fracture detection and assessing the efficacy of AI models and their impact on performance purposes. This study evaluated two implication phases of AI models amongst 3,016 paediatric radiographs and 189 cases, overall finding that the AI model incorporated into paediatric fracture tests exhibited high accuracy in detecting fractures, and its integration significantly enhanced reader performance.
Both of these studies exhibit AI assistance in bone fraction detection in patients, allowing for quicker turnaround time, and enhancing radiologists’ reader performance.
