From: A holistic approach to implementing artificial intelligence in radiology
Type of AI application | State of adoption/integration | Impact on clinical practice |
---|---|---|
3D tumour segmentation of vestibular schwannoma on MRI | Ready for prospective validation in research setting | Transition from 2D measurements to automated 3D volume measurement, resulting in time reduction and quality improvement |
Normal/abnormal detection chest X-ray | Fully integrated into clinical workflow (validated on prospective data). AI output directly present in worklist of radiologist. Detailed overlay images presented in separated viewer | ± 45% of all chest X-ray cases in clinical practice are normal. The algorithm is able to automate reporting for approximately 20% of all normal cases, enhancing clinical efficiency |
Lung nodule detection on CT-Thorax | Fully integrated into clinical workflow (validated on prospective data). Output (number of detected lung nodules and percentage of affected lung tissue) presented in PACS worklist and possibility to accept/reject/modify nodule segmentations in PACS viewer | Substantial time reduction in follow-up imaging and improved lung nodule comparisons over time |
Bone age measurements on X-ray | Fully integrated into clinical workflow (validated on prospective data). AI report automatically available in PACS | Automated bone age measurements on x-ray, facilitating task differentiation to advanced practitioners |
Covid detection and quantification on CT | Fully integrated into clinical workflow (validated on prospective data). AI report automatically available in PACS | Robust quantification of COVID-affected lung parenchyma in all lung segments, significantly enhancing reporting efficiency and quality |
Leg angle and distance measurements on X-ray | Fully integrated into clinical workflow (validated on prospective data). AI report is automatically available in PACS. Standardised radiological report based on AI output | Automated leg angle and distance measurements with an AI acceptance rate of approximately 90% |
MRI neuro quantification for dementia patients | Fully integrated into clinical workflow (validated on prospective data). AI report automatically available in PACS | Automated quantification of white matter abnormalities and atrophy evaluation |
Automatic quality feedback for chest X-ray | Fully integrated in clinical workflow. Automatic quality feedback on iPad after image acquisition | Enhancement of image quality to ensure accurate reporting and prevent the need for patients to return, as low-quality images may otherwise necessitate their return for re-imaging |
Large vessel occlusion detection for early stroke detection | Fully integrated into clinical workflow (validated on prospective data) | |
Fracture detection on X-ray | Implementation phase. Connected to clinical systems and ready for clinical use | Potential impact: decreased reporting time, enhanced diagnostic confidence and subsequently boost job satisfaction, particularly during night and weekend shifts when residents work independently |
Scoliosis measurements on X-ray | Implementation phase. Connected to clinical systems and ready for clinical use | Potential impact: automated scoliosis measurements |
Automated vertebral fracture assessment on DXA | Development phase, model development and retrospective validation | Potential impact: automated vertebral fracture assessments resulting in significant reduction in reporting time. Prototype has shown positive impact on reader discomfort for annotation |
Prostate analysis on MRI | Exploration phase | Potential impact: decreased reporting time by pre-filled structured report based on AI output |
AI for tomosynthesis | Exploration phase | Potential impact: decreased reporting time by pre-filled structured report based on AI output |
Knee osteoarthritis measurements on X-ray | Exploration phase | Potential impact: automated osteoarthritis measurements. Reduction in reporting time |