Integration of detection and tracking networks for automated rib multiplanar reconstruction: a feasibility study for fracture diagnosis.
Authors
Affiliations (4)
Affiliations (4)
- Department of Radiology, Tongji Hospital of Tongji Medical College of Huazhong University of Science and Technology, Wuhan, China.
- School of Mechanical Science and Engineering, Huazhong University of Science and Technology, Wuhan, China.
- Department of Radiology, Tongji Hospital of Tongji Medical College of Huazhong University of Science and Technology, Wuhan, China. [email protected].
- School of Mechanical Science and Engineering, Huazhong University of Science and Technology, Wuhan, China. [email protected].
Abstract
Rib fractures are common yet time-consuming to diagnose. This study explores automation via multiplanar reconstruction and intelligent detection algorithms to accelerate and optimize clinical assessment. A retrospective study was conducted with data from consecutive three-dimensional (3D) computed tomography (CT) examinations of the ribs in 230 patients (137 males), aged 51.7 ± 13.0 years (mean ± standard deviation). Object detection with tracking algorithms and integrated evaluation functions was applied to construct the automatic multiplanar reconstruction (MPR) system. Two readers independently conducted evaluations using automatic multiplanar reconstructions, curved surface reconstructions (CSR), and 3D reconstructions. Results were compared to a reference standard (RS) created by two senior radiologists. Of 5,520 ribs analyzed, 1,065 (19.3%) were positive at RS. Using automatic MPR, overall 85.4% sensitivity (910/1,065) (95% confidence interval 83.2‒87.5%) and 98.9% specificity (4,406/4,445) (95% CI: 98.5‒99.2%) were obtained. The performance of original CT, CSR, and 3D images was: sensitivity 94.2%, 79.4%, and 58.2%; and specificities 99.6%, 96.2%, and 99.2%, respectively. Reading time decreased by approximately 75% from 159.3 ± 50.5 s using original CT images to 41.2 ± 6.6 s using MPR. The automatic MPR system offered an accurate solution for diagnosing rib lesions, reducing the reading time. While superior to CSR and 3D reconstructions, automatic MPR should be regarded as a complement to, rather than a substitute for, original CT images in its current form. Future research expanding datasets, exploring different clinical scenarios, and enhancing training for younger physicians is expected. Automatic MPR significantly improves rib fracture diagnosis speed and accuracy, reducing evaluation time by 75%. This artificial intelligence system enhances radiologist performance and promises broader clinical integration in trauma care and emergency imaging workflows. Over 5,500 ribs were analyzed, with 1,065 (19.3%) positive at the reference standard created by two senior radiologists. Using automatic MPR, overall 85.4% sensitivity and 98.9% specificity were obtained. The reading time decreased by approximately 75% from 159.3 ± 50.5 s using original CT images to 41.2 ± 6.6 s using MPR.