Detecting pelvic fracture on 3D-CT using deep convolutional neural networks with multi-orientated slab images

Abstract

Pelvic fracture is one of the leading causes of death in the elderly, carrying a high risk of death within 1 year of fracture. This study proposes an automated method to detect pelvic fractures on 3-dimensional computed tomography (3D-CT). Deep convolutional neural networks (DCNNs) have been used for lesion detection on 2D and 3D medical images. However, training a DCNN directly using 3D images is complicated, computationally costly, and requires large amounts of training data. We propose a method that evaluates multiple, 2D, real-time object detection systems (YOLOv3 models) in parallel, in which each YOLOv3 model is trained using differently orientated 2D slab images reconstructed from 3D-CT. We assume that an appropriate reconstruction orientation would exist to optimally characterize image features of bone fractures on 3D-CT. Multiple YOLOv3 models in parallel detect 2D fracture candidates in different orientations simultaneously. The 3D fracture region is then obtained by integrating the 2D fracture candidates. The proposed method was validated in 93 subjects with bone fractures. Area under the curve (AUC) was 0.824, with 0.805 recall and 0.907 precision. The AUC with a single orientation was 0.652. This method was then applied to 112 subjects without bone fractures to evaluate over-detection. The proposed method successfully detected no bone fractures in all except 4 non-fracture subjects (96.4%).

Automated fracture screening using an object detection algorithm on whole-body trauma computed tomography

“…Additionally, these authors examined whether a CNN model could achieve accurate anatomical localization (right 1st–12th and left 1st–12th ribs) and classification (fresh, healing, and old fractures) of rib fractures, and showed that the sensitivity reached 0.971 and 0.949 on the right and left ribs, respectively 20 . Ukai et al 17 proposed an automated method to detect pelvic fractures in the 3D fracture region obtained by integrating multiple 2D fracture candidates, and reported that sensitivity was 0.805 and precision was 0.907. The ability of the CNN model in our study to automatically localize and classify fractures in whole-body CT axis slices of pelvic, rib, and spine fractures was comparable to that of previous CT studies.…”

“…New CNN algorithms reduce human workload and can extract features that are difficult for humans to recognize. Some studies have reported fracture identification using CNN on radiographs and CT scans 15 – 17 . However, to our knowledge, automatic localization and classification of fractures in CT using CNN methods have only been reported for rib and pelvic fractures and each model can detect only a single type of fracture 16 , 17 .…”

Effect of microstructural heterogeneity on fatigue strength predicted by reinforcement machine learning

“…Surface fractures may not be seen in raw images. 44 These fracture properties and their combinations properly depict surface fatigue. 43 In practice, the time-consuming magnetic particle imaging (MPI) inspection approach does not apply to lowautomation railway infrastructure.…”

“…The scans were preprocessed using image processing algorithms to identify characteristics similar to surface cracks. Surface fractures may not be seen in raw images 44 . These fracture properties and their combinations properly depict surface fatigue 43 .…”

Artificial intelligence fracture recognition on computed tomography: review of literature and recommendations

“…All seventeen studies used a CNN to detect and /or classify fractures on CT scans [12][13][14][15][16][17][18][19][20][21][22][23][24][25][26][27][28]. Eight studies addressed detection of rib fractures [13,17,19,20,22,[25][26][27], three studies the performance for detection [12,21] and classification [18] of pelvic fractures, four for detection of spine fractures [14,16,23,28], one for detection and classification of femur fractures [24] and one of calcaneal fractures [15]. Fourteen studies used two output classes (fracture yes/no).…”