Study (Publication year) | Study period | Type of image | Type of fracture | Number of images | Fracture classification | Reference diagnosis of fracture diagnosis and classification | Augments |
---|---|---|---|---|---|---|---|
Adams [11] | X | X-ray | Normal, Neck Fx | 805 images | Fracture (n = 403) No fracture (n = 402) | 142 undergraduate students for the detection of neck of femur fractures" | MATLAB’s inbuilt image database augmentation algorithms |
Urakawa  [12] | 2006.1–2017.7 | X-ray | Normal, ITC Fx | 3346 images from 1773 patients | Fracture (n = 1773) No fracture (n = 1573) | Single boardcertified orthopedic surgeon using a Digital Imaging and Communications in Medicine viewer | 2650 iterations, or training of 132,500 (2650 × 50) augmented images using the adaptive moment estimation (Adam) optimizer |
Cheng [13] | 2008.8–2016.12 | X-ray | Normal, hip Fx | 3605 images | Fracture (n = 1975) No fracture (n = 1630) | The radiologist’s report, diagnosis, clinical course, and other related images, such as views of the hip joint, were reviewed |  |
Krogue [14] | 1998–2017 | X-ray | Normal, hip Fx | 3026 images from 3026 patients | Nondisplaced femoral neck fractures (n = 182) Displaced FN fractures (n = 525) Intertrochanteric fractures (n = 765) No fracture (n = 1554) including Arthroplasty (n = 172) and Open reduction internal fixation (n = 59) | Two postgraduate year 4 orthopedic residents using the Visual Geometry Group Image Annotator | Three types of contrast changing cut-out, Gaussian-mixture masking, and bounding box wiggling |
Yu [15] | X | X-ray | Normal, hip Fx | 1061 images from 617 patients | Subcapital or transcervical fracture (n = 185) Basicervical or intertrochanteric fracture (n = 216) Subtrochanteric fractures (n = 50) No fracture (n = 610) | Local experts | X |
Mutasa [16] | 2000.2–2017.2 | X-ray | Normal, Neck Fx | 1063 images from 550 patients | Garden I/II fracture (n = 127) Garden III/IV fracture (n = 610) No fracture (n = 326) | Musculoskeletal fellowship-trained radiologists | (1) 1063 Source images modification including image flipping, random rotation, and random contrast jittering (2) 6000 digitally reconstructed radiographs (DRRs) were generated using simulated X-ray volume rendering (3) Additional 2000 training examples were generated utilizing a generative adversarial network (GAN) |
Beyaz [17] | 2013.1–2018.1 | X-ray | Normal, Neck Fx | 234 images from 65 patients | Fracture (n = 149) No fracture (n = 85) | X | 2106 Source images modification including rotation, and Gaussian noise |
Mawatari [18] | 2004.4–2018.4 | X-ray, CT, MRI | Normal, hip Fx | 352 images | Fracture (n = 327) No fracture (n = 25) | 2 radiologists | Image rotation with the allowance of plus and minus one degree, and horizontal flipping. (n = 3300) |
Yamada [19] | 2014.3–2020.1 | X-ray | Normal, hip Fx | 2,923 images (1,703 anteriposterior view, 1220 lateral view) from 1035 patients | Fracture (n = 1983) No fracture (n = 940) | 2 board-certified orthopedic surgeons | Rotation angle range of 20°, width shift range of 0.2, height shift range of 0.2, brightness range of 0.3–1.0, and a horizontal flip of 50% using ImageDataGenerator |
Cheng [20] | 2008.8–2016.12 | X-ray | Normal, hip Fx | 3605 images, 587 Real data | Fracture (n = 1975) No fracture (n = 1630) | Based on all the available clinical information, including clinical diagnosis, imaging reports, advanced imaging reports, and operative findings" | X |
Yoon [8] | 2016–2018 | X-ray, CT | Normal, ITC Fx | 3343 images | Two classes: no fracture and fracture Three classes: no fracture, A1.1 to A2.1, and A2.2 to A3.3 Four classes: no fracture, A1, A2, and A3 Seven classes: no fracture, each type of A1.1 to A1.3, each type of A2.1 to A2.3, and A3 Ten classes: no fracture, each type of A1.1 to A1.3, each type of A2.1 to A2.3, and each type of A3.1 to A3.3 Each classes are according to AO/OTA classification [1] | Orthopedic surgeons | X |
Sato [1] | 2009–2019 | X-ray | Normal, hip Fx | 10,484 images from 4,851 patients | Fracture (n = 5242) No fracture (n = 4851) | 2 orthopedic surgeons | X |
Bae [21] | 2005.1–2018.12 | X-ray | Normal, Neck Fx | 4189 images | Fracture (n = 1109) No fracture (n = 3080) | 2 emergency medicine specialists | Random transformation including flip, flop, or rotation |
Murphy [7] | X | X-ray | Normal, hip Fx | 3659 images | Intracapsular fracture (n = 1082) Trochanteric fracture (n = 974) No fracture (n = 1603) | 2 musculoskeletal experts (consultant orthopedic surgeon and/or consultant musculoskeletal radiologist) | Random rotating the images (−10° and 10°), grayscale-inverted (chosen at random), and using mirrored images. (n = 47,698) |