Lightweight Brain Tumor Segmentation Using Semantic Flow and Scale Perception

Table 1. Datasets composition
View table
Table 1. Datasets composition
Dataset Data quantity in train dataset Data quantity in valid dataset
BraTS2020 369 128
BraTS2019 335 125
BraTS2018 285 66

Table 2. Ablation experiment of SAHNet model

View table

Table 2. Ablation experiment of SAHNet model

Model	Dice /%			Sensitivity /%			Specificity /%			95% Hausdorff distance /mm
Model	ET	WT	TC	ET	WT	TC	ET	WT	TC	ET	WT	TC
HDC^［24］	77.18	89.14	81.21	76.37	88.52	77.64	99.97	99.14	99.96	26.90	9.21	12.83
HDC+SAM	78.30	89.31	81.02	78.11	89.49	79.97	99.97	99.90	99.95	15.86	5.23	18.24
HDC+FA	77.04	89.43	82.09	77.46	90.29	81.15	99.97	99.90	99.95	20.85	7.19	11.83
HDC+MHDC	77.88	89.55	80.07	77.68	90.46	78.03	99.97	99.90	99.96	21.38	6.53	15.63
HDC+MHDC+FA	78.01	89.63	80.47	78.92	90.83	77.94	99.97	99.90	99.94	24.23	5.04	12.29
HDC+SAM+FA	77.88	89.47	82.02	77.65	89.49	79.24	99.97	99.91	99.65	24.32	5.80	14.46
HDC+SAM+MHDC	78.03	89.76	80.82	77.93	91.33	78.82	99.97	99.89	99.96	23.64	5.12	9.59
SAHNet	77.61	89.50	83.13	78.43	90.38	81.21	99.97	99.90	99.96	26.57	9.19	8.38

Table 3. Comparative experimental results between SAHNet and different models

View table

Table 3. Comparative experimental results between SAHNet and different models

Type	Model	Dice /%			95% Hausdorff distance /mm			N_Para/ 10⁹	FLOPs / 10⁹
Type	Model	ET	WT	TC	ET	WT	TC	N_Para/ 10⁹	FLOPs / 10⁹
Classic network	3DUNet^［29］	76.26	87.71	79.37	29.83	6.79	24.23	16.12	176.61
	V-NET^［16］	68.97	86.11	77.90	45.51	14.49	16.15
	SAHNet （ours）	77.61	89.50	83.13	26.57	9.19	6.38	0.52	41.46
Popular network	TransBTS （2021）^［20］	78.73	90.09	81.73	17.94	4.69	9.76	15.14	208.00
	SwinBTS （2022）^［30］	77.36	89.06	80.30	26.84	17.06	15.78	20.40
	dResUnet （2023）^［31］	80.04	86.60	83.51				30.47	374.04
	ASTNet （2022）^［32］	77.82	90.40	84.21	17.04	4.69	9.13	10.05	44.1
	SAHNet （ours）	77.61	89.50	83.13	26.57	9.19	6.38	0.52	41.46
Lightweight network	AD-Net （2023）^［33］	76.49	90.11	80.53	35.45	7.22	15.31	5.27	134.98
	DMF （2019）^［23］	76.46	88.78	80.43	20.69	8.82	12.379	3.88	27.04
	HDC （2020）^［24］	77.18	89.14	81.21	26.90	9.21	12.83	0.31	24.00
	HMNet （2023）^［34］	78.10	90.07	81.20	21.34	5.95	7.05	0.80	129.4
	SAHNet （ours）	77.61	89.50	83.13	26.57	9.19	6.38	0.52	41.46

Table 4. Comparison of SAHNet with other lightweight networks on different datasets

View table

Table 4. Comparison of SAHNet with other lightweight networks on different datasets

Dataset	Model	Dice /%			95% Hausdorff distance /mm			N_Para /10⁶	FLOPs /10⁹
Dataset	Model	ET	WT	TC	ET	WT	TC	N_Para /10⁶	FLOPs /10⁹
BraTS2018	DMF （2019）^［23］	80.11	90.61	84.54	3.06	4.66	6.44	3.88	27.04
	HDC （2020）^［24］	80.13	89.80	84.59	2.39	6.86	5.71	0.31	24.00
	HMNet （2023）^［33］	78.63	90.10	84.32	2.69	4.72	7.73	0.80	129.40
	DAFANet （2023）^［35］	80.44	90.07	84.54	3.07	4.36	6.23	4.23	30.50
	SAHNet （ours）	81.76	90.30	85.44	2.37	4.70	5.31	0.52	41.46
BraTS2019	AD-Net （2023）^［32］	76.31	90.16	81.51	35.50	4.31	13.42	5.27	134.98
	DMF （2019）^［23］	76.68	88.47	80.19	3.34	5.41	6.14	3.88	27.04
	HDC （2020） ^［24］	76.98	89.15	81.51	4.15	6.43	8.01	0.31	24.00
	HMNet （2023）^［33］	77.23	89.93	83.01	4.01	5.21	6.57	0.80	129.40
	DAFANet （2023）^［35］	78.11	89.15	81.51	3.07	4.36	6.22	4.23	30.50
	SAHNet （ours）	78.09	89.94	83.26	3.21	4.93	5.56	0.52	41.46

Tools

Get Citation

Copy Citation Text

Chuanqiang Liu, Xiaoqi Lü, Jing Li, Yu Gu. Lightweight Brain Tumor Segmentation Using Semantic Flow and Scale Perception[J]. Chinese Journal of Lasers, 2025, 52(3): 0307105

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites