Lightweight Dental Image Segmentation with Quadrant Oblique Displacement

Fig. 5. Comparisons of dental segmentation results of different models on STS MICCAI 2023 dataset. (a) Input images; (b) ground truth; (c) UNet; (d) UNet++; (e) ResU-Net; (f) CaraNet; (g) TransUNet; (h) MedT; (i) UNeXt; (j) QOD-UNeXt

Download full size

Fig. 6. Comparisons of dental segmentation results of different models on Tufts dataset. (a) Input images; (b) ground truth; (c) UNet; (d) UNet++; (e) ResU-Net; (f) CaraNet; (g) TransUNet; (h) MedT; (i) UNeXt; (j) QOD-UNeXt

Download full size

Table 1. Overview of datasets used for dental image segmentation
View table
Table 1. Overview of datasets used for dental image segmentation
Dataset name Number of images Original resolution Processed resolution
STS-MICCAI 2023 2900 640×320 256×256
Tufts 1000 1615×840 256×256

Table 2. Comparison of parameters, GFLOPs, and inference time among different models

View table

Table 2. Comparison of parameters, GFLOPs, and inference time among different models

Type	Model	Parameters /M	GFLOPs /G	Inference time /ms
CNN	UNet^［4］ UNet++^［7］ ResU-Net^［5］ CaraNet^［8］	32.52 48.99 13.04 24.92	107.07 575.61 809.84 59.52	18 53 70 20
Transformer	TransUNet^［14］ MedT^［16］	58.68 1.37	305.44 24.06	38 292
Lightweight	UNeXt^［28］	1.47	5.73	6
Lightweight	LightM-UNet^［27］	2.96	238.53	69
Mamba	U-Mamba^［19］	175.87	1803.00	171
KANs	U-KAN^［18］	9.38	27.55	56
Ours	QOD-UNeXt	1.35	7.68	11

Table 3. Comparisons of results among different models on STS-MICCAI 2023 dataset

View table

Table 3. Comparisons of results among different models on STS-MICCAI 2023 dataset

Type	Model	STS-MICCAI 2023 dataset
Type	Model	Dice /%	IoU /%	SEN /%	SPE /%
CNN	UNet^［4］ UNet++^［7］ ResU-Net^［5］ CaraNet^［8］	79.40 ± 0.44 86.31 ± 0.49 86.27 ± 0.59 87.60 ± 0.55	71.61 ± 0.48 76.82 ± 0.69 76.59 ± 0.75 78.79 ± 0.69	76.71 ± 0.55 86.70 ± 0.41 85.97 ± 0.83 88.11 ± 0.75	91.56 ± 0.73 95.80 ± 0.80 94.10 ± 0.52 93.42 ± 0.20
Transformer	TransUNet^［14］ MedT^［16］	89.82 ± 0.47 87.80 ± 0.28	81.79 ± 0.26 78.86 ± 0.37	87.81 ± 0.53 89.13 ± 0.12	96.81 ± 0.70 92.73 ± 0.94
Lightweight	UNeXt^［28］ LightM-UNet^［27］	88.84 ± 0.68 87.15 ± 0.52	80.98 ± 0.39 79.94 ± 0.43	88.40 ± 0.86 87.05 ± 0.62	95.34 ± 0.03 94.62 ± 0.58
Mamba	U-Mamba^［19］	90.47 ± 0.45	83.49 ± 0.54	89.32 ± 0.71	97.89 ± 0.67
KANs	U-KAN^［18］	89.68 ± 0.39	81.73 ± 0.49	89.47 ± 0.65	96.12 ± 0.72
Ours	QOD-UNeXt	92.35 ± 0.40	86.00 ± 0.70	91.80 ± 0.11	98.57 ± 0.35

Table 4. Comparisons of results among different models on Tufts dataset

View table

Table 4. Comparisons of results among different models on Tufts dataset

Type	Model	Tufts dataset
Type	Model	Dice /%	IoU /%	SEN /%	SPE /%
CNN	UNet^［4］ UNet++^［7］ ResU-Net^［5］ CaraNet^［8］	78.39 ± 0.60 81.11 ± 0.52 80.23 ± 0.02 82.33 ± 0.27	71.20 ± 0.66 73.09 ± 0.58 69.52 ± 0.84 70.95 ± 0.20	87.85 ± 0.10 90.98 ± 0.65 81.73 ± 0.99 90.22 ± 0.54	89.49 ± 0.08 88.99 ± 0.35 90.94 ± 0.56 87.59 ± 0.26
Transformer	TransUNet^［14］ MedT^［16］	83.71 ± 0.11 82.23 ± 0.95	72.97 ± 0.80 71.87 ± 0.02	89.19 ± 0.02 91.74 ± 0.40	91.81 ± 0.50 92.63 ± 0.82
Lightweight	UNeXt^［28］ LightM-UNet^［27］	82.69 ± 0.65 82.15 ± 0.58	72.72 ± 0.84 74.20 ± 0.76	88.08 ± 0.46 89.65 ± 0.55	95.07 ± 0.23 93.80 ± 0.34
Mamba	U-Mamba^［19］	87.12 ± 0.48	78.20 ± 0.71	92.35 ± 0.62	95.50 ± 0.28
KANs	U-KAN^［18］	86.43 ± 0.42	77.65 ± 0.69	91.60 ± 0.53	94.70 ± 0.32
Ours	QOD-UNeXt	90.32 ± 0.30	82.86 ± 0.40	93.43 ± 0.81	97.24 ± 0.06

Table 5. Results of ablation study on STS-MICCAI 2023 and Tufts dataset

View table

Table 5. Results of ablation study on STS-MICCAI 2023 and Tufts dataset

Method	STS-MICCAI 2023 dataset				Tufts dataset
Method	Dice /%	IoU /%	SEN /%	SPE /%	Dice /%	IoU /%	SEN /%	SPE /%
Experiment A	84.21	77.94	80.04	87.66	82.06	74.40	84.14	87.97
Experiment B	88.58	82.27	87.23	93.95	87.13	78.91	89.16	92.59
Experiment C	90.05	83.45	89.72	95.74	88.21	80.37	90.56	94.38
Experiment D	87.28	80.72	86.15	92.12	86.02	77.42	87.29	90.73
Experiment E	83.91	77.39	79.96	86.83	81.95	73.91	83.27	87.11
Experiment F	82.44	76.21	77.47	85.04	80.87	72.45	81.87	85.32
QOD-UNeXt	92.35	86.00	91.80	98.57	90.32	82.86	93.43	97.24

Tools

Get Citation

Copy Citation Text

Ziyuan Yin, Yun Wu. Lightweight Dental Image Segmentation with Quadrant Oblique Displacement[J]. Laser & Optoelectronics Progress, 2025, 62(10): 1037007

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites