notesum.ai

Published at December 4

Intuitive Axial Augmentation Using Polar-Sine-Based Piecewise Distortion for Medical Slice-Wise Segmentation

cs.CV

cs.AI

Released Date: December 4, 2024

Authors: Yiqin Zhang¹, Qingkui Chen¹, Chen Huang², Zhengjie Zhang¹, Meiling Chen¹, Zhibing Fu¹

Aff.: ¹University of Shanghai for Science and Technology; ²Shanghai General Hospital

Arxiv: http://arxiv.org/pdf/2412.03352v1

Refer to caption

Model	Implementation	Metric	Classes						Average
Model	Implementation	Metric	backgound	bladder	bone	kidney	lung	liver	Average
MedNext	Vanilla	Dice	96.39	65.53	3.21	19.32	66.26	86.13	56.14
		Precision	96.01	62.91	12.70	76.04	67.88	78.92	65.74
		Recall	96.77	68.37	1.84	11.07	64.72	94.80	56.26
	With Ours	Dice	96.85	78.54	48.55	36.08	71.19	90.04	70.21
		Precision	95.60	79.93	88.49	78.35	82.75	89.58	85.78
		Recall	98.13	77.19	33.45	23.44	62.47	90.50	64.20
	Average Improve		+0.47	+12.95	+50.91	+10.48	+5.85	+3.42	+14.01
SwinUMamba	Vanilla	Dice	96.85	71.44	1.39	21.38	71.84	92.3	59.21
		Precision	95.27	80.27	47.50	70.38	78.02	91.40	77.14
		Recall	98.49	64.36	0.71	12.61	66.56	93.34	56.01
	With Ours	Dice	97.27	74.52	58.87	32.81	73.57	90.23	71.21
		Precision	96.04	73.63	82.58	70.40	86.26	91.52	83.41
		Recall	98.53	75.44	45.74	21.39	64.13	88.98	65.70
	Average Improve		+0.41	+2.51	+45.86	+6.74	+2.51	-2.12	+9.32