Table 1.

Performance of various models for fracture detection.

No.	Author	Year	Modality	Model/Method	Skeletal Joints	Description	Performance
1	Kim et al. [30]	2018	Xray/MRI	Inception V3	Wrist	The author proved that the concept of transfer learning from CNNs in fracture detection on radiographs can provide the state of the performance.	AUC = 0.954 Sensitivity = 0.90 Specificity = 0.88
2	Olczak et al. [31]	2017	Xray/MRI	BVLC Reference CaffeNet network/VGG CNN/Network-in- network/VGG CNN S	Various Parts	Here, the research supports the use of deep learning to outperform the human performance.	Accuracy = 0.83
3	Cheng et al. [33]	2019	Radiographic images	DenseNet 121	Hips	The aim of this study was to localise and classify hip fractures using deep learning.	$\mathrm{Accuracy} \approx 91$ $\mathrm{Sensitivity} \approx 98$ $\mathrm{Specificity} \approx$ 84
4	Chung et al. [32]	2018	Radiographic images	ResNet 152	Humeral	The authors proposed a model for the detection and classification of the fractures from AP shoulder radiographic images.	$\mathrm{Accuracy} \approx 96$ $\mathrm{Sensitivity} \approx 0.99$ $\mathrm{Specificity} \approx 0.97$ $\mathrm{AUC} \approx$ 0.996
5	Urakawa et al. [10]	2018	Radiographic images	VGG_16	Hips	This study shows a comparison of diagnostic performance between CNNs and orthopaedic doctors.	$\mathrm{Accuracy} \approx 95.5$ $\mathrm{Sensitivity} \approx 93.9$ $\mathrm{Specificity} \approx 97.40$ $\mathrm{AUC} \approx$ 0.984
6	Kitamura et al. [36]	2019	Radiographic images	7 modelsInception V3 ResNet (with/without drop&aux) Xception (with/without drop&aux) Ensemble A Ensemble B	Ankle	The study was done in order to determine the efficiency of CNNs on small datasets.	Best performance by Ensemble_A $\mathrm{Accuracy} \approx 83$ $\mathrm{Sensitivity} \approx 80$ $\mathrm{Specificity} \approx$ 81
7	Yu [41]	2020	Radiographic images	Inception V3	hip	The proposed algorithm performed well in terms of APFF detection, but not so well in terms of fracture localization.	Accuracy = 96.9 AUC = 0.994 Sensitivity = 97.1 Specificity = 96.7
8	Gan [42]	2019	Radiographic images	Inception V4	Wrist	The authors implemented the algorithm for the detection of distal radius fractures.	Accuracy = 93 AUC = 0.961 Sensitivity = 90 Specificity = 96
9	Choi [43]	2019	Radiographic images	ResNet 50	Elbow	The authors aimed the development of dual input CNN-based deep learning model for automated detection of supracondylar fracture.	AUC = 0.985 Sensitivity = 93.9 Specificity = 92.2
10	Majkowska et al. [44]	2020	Radiographic images	Xception	Chest	The authors developed a model to detect opacity, pneumothorax, mass or nodule, and fracture.	$\mathrm{AUC} \approx$0.86 Sensitivity = 59.9 Specificity = 99.4
11	Lindsey et al. [8]	2018	Radiographic images	Unet	wrist	This study involves the implementation of deep learning to help doctors to distinguish between fractured and normal wrist.	AUC = 97.5% Sensitivity = 93.9% Specificity = 94.5
12	Johari et al. [45]	2016	Radiographic images	probabilistic neural network (PNN) CBCT-G1/2/3, PA-G1/2/3	Vertical Roots	This study supports the initial detection of vertical roots fractures.	Best performance by PNN Model $\mathrm{Accuracy} \approx 96.6$ $\mathrm{Sensitivity} \approx 93.3$ $\mathrm{Specificity} \approx$ 100
13	Heimer et al. [46]	2018	CT	deep neural networks.	Skull	The study aims at classification and detection of skull fractures curved maximum intensity projections (CMIP) using deep neural networks.	CMPIs THRESHOLD = 0.79 Specificity= 87.5 Sensitivity =91.4 CMPIs THRESHOLD = 0.75 Specificity= 72.5 Sensitivity =100
14	Wang et al. [11]	2022	CT	CNN	Mandibule	The author implemented a novel method for the classification and detection of mandibular fracture.	Accuracy = 90% AUC = 0.956
15	Rayan et al. [9]	2021	Radiographic images	XceptionNet	elbow	This study aims for a binomial classification of acute paediatric elbow radiographic abnormalities.	AUC = 0.95 Accuracy = 88% Sensitivity = 91% Specificity = 84%
16	Adam et al. [34]	2019	Radiographic images	AlexNet and GoogLeNet	femur	Here, the author aimed to evaluate the accuracy of DCNN for the detection of femur fractures.	Accuracy AlexNet = 89.4% GoogLeNet = 94.4%
17	Balaji et al. [35]	2019	x-ray	CNN based model	Diaphyseal Femur	In this study, the author implemented an automated detection and diagnosis of femur fracture.	Accuracy = 90.7% Specificity = 92.3% Sensitivity = 86.6%
18	Pranata et al. [37]	2020	Radiographic images	convolutional neural network (CNN)	Femoral neck	In this study, the author aimed at the detection of femoral neck fracture using genetic and deep learning methods.	Accuracy = 0.793 Specificity = 0.729 Sensitivity = 0.829
19	Rahmaniar et al. [26]	2019	CT	Computerised system	Calcaneal fractures	Here, the author aims at automated segmentation and detection of calcaneal fractures.	Accuracy = 0.86
20	Burns et al. [47]	2017	CT	Computerised system	spine	The author implemented a computerized system to detect classify and localize compression fractures.
21	Tomita et al. [39]	2018	CT	Deep convolutional neural network	vertebra	This study aims at the early detection of osteoporotic vertebral fractures.	Accuracy = 89.2% F1 score = 90.8%
22	Muehlematter et al. [38]	2018	CT	Machine-learning algorithms	vertebra	Here, the author aims at evaluation of the performance of bone texture analysis with a machine learning algorithm.	AUC = 0.64