Table 3 Comparison of the proposed work with the studies.

From: An efficient deep learning based approach for automated identification of cervical vertebrae fracture as a clinical support aid

Author

No of Sample

Accuracy

Sensitivity

Specificity

Comments

Sha et al17.

7836

98.6%

83.8%

97.9%

U-Net with dilated convolution and attention for spinal lesion segmentation in CT, improving diagnostic speed.

Shim et al18.

707

99.13%

90.44%

99.51%

Compared U-Net variations, Attention U-Net performed best for cervical spine segmentation in X-rays.

Sha et al19.

N/A

87.9%

N/A

N/A

U-Net with preprocessing and data augmentation for spinal fracture segmentation in clinical use.

Bae et al20.​

1684 disease slices, 3490 healthy slices

96.23%

N/A

N/A

Fully automated 2D U-Net for 3D cervical vertebra segmentation, achieved high DSC scores in validation.

Paul et al21. ​

4200

99.75%

N/A

N/A

MobileNetV2 model achieved highest accuracy for cervical spine fracture classification in CT images.

Xu et al22.​

CTSpine1K, VerSe 20

88.4% (CTSpine1K)

N/A

N/A

Residual U-Net combined with Transformer for vertebral edge feature fusion, state-of-the-art segmentation.

A. Dastgir et al35.

10,000 X-ray images

(VinDr-SpineXR)

96.81%

97.95%

N/A

MAFMv3: MobileNetV3 + CBAM + ASPP + histogram-equalized views for spine lesion classification

M.U. Saeed et al36.

80 CT test samples

(VerSe19/20)

N/A

96.52% (VerSe20)

94.64% (VerSe19)

N/A

3D MFA (Multi-Feature Attention): MobileNetV3 + Reverse CBAM + FPP + ASPP; lightweight and pruned architecture.

Salehinejad et al13.

3666

79.18%

57 — 64%

77 — 84%

Uses a bidirectional long-short term memory (BLSTM) layer on CT imaging for the cervical spine fracture detection

J.E. Small et al14.

665

92% (CNN)

95%

(radiologists)

76% (CNN)

93% (radiologists)

97% (CNN)

96% (radiologists)

CNN was designed to detect cervical spine fractures on CT and compared it to that of radiologists.

Arunnit Boonrod et al15.

625

75%

80%

72%

Trained different variants of YOLO network (v2, v3, and v4). Among these, YOLO v4 used to detect cervical spine injury.

Presented Work

29,832 CT-scan slices

98.44%

92 — 97%

90 — 98%

Inception-ResNet-v2 U-Net (fine-tuned) used to detect the spine fracture.

  1. N/A — Not Available
Back to article page