notesum.ai

Published at November 27

PDZSeg: Adapting the Foundation Model for Dissection Zone Segmentation with Visual Prompts in Robot-assisted Endoscopic Submucosal Dissection

cs.CV

cs.AI

Released Date: November 27, 2024

Authors: Mengya Xu¹, Wenjin Mo², Guankun Wang¹, Huxin Gao¹, An Wang¹, Zhen Li³, Xiaoxiao Yang³, Hongliang Ren¹

Aff.: ¹Dept. of Electronic Engineering, CUHK, Hong Kong, China.; ²Dept. of Computer Science and Engineering, Sun Yat-sen University, Guangzhou, China.; ³Dept. of Gastroenterology, Qilu Hospital of Shandong University, Jinan, China.

Arxiv: http://arxiv.org/abs/2411.18169v1

		Dissection zone		No-go zone
Visual Prompt	Model	IoU	Dice	IoU	Dice	Mean IoU	Mean Dice
	SETR	43.91	61.02	97.93	98.95	70.92	79.99
	STDC	41.92	59.07	97.79	98.88	77.04	78.97
	Point-Rend	44.71	61.79	97.65	98.81	71.18	80.30
	Fast-SCNN	41.86	59.01	97.77	98.87	69.81	78.94
	DeepLabv3	45.95	62.97	97.90	98.94	71.93	80.95
Without visual prompt	Our DZSeg	47.55	63.01	98.05	99.04	72.80	81.02
	SETR	50.53	67.14	98.18	99.08	74.36	83.11
	STDC	51.64	68.11	98.21	99.10	74.93	83.61
	Point-Rend	54.49	70.54	98.25	99.12	76.37	84.83
	Fast-SCNN	51.51	68.00	98.25	99.12	74.88	83.56
	DeepLabv3	55.57	71.44	98.37	99.18	76.97	85.31
	SAM-Adapter	41.43	55.44	/	/	/	/
With point visual prompt	Our PDZSeg	56.37	71.43	98.46	99.22	77.41	85.32
	SETR	64.48	78.4	98.75	99.37	81.61	88.89
	STDC	69.41	81.94	98.90	99.45	84.15	90.69
	Point-Rend	67.87	80.86	98.80	99.40	83.33	90.13
	Fast-SCNN	66.36	79.78	98.73	99.36	82.55	89.57
	DeepLabv3	69.24	81.83	98.90	99.45	84.07	90.64
With bounding box visual prompt	Our PDZSeg	69.85	82.00	98.97	99.48	84.41	90.73
	SETR	62.38	76.83	98.66	99.33	80.52	88.08
	STDC	62.65	77.04	98.58	99.28	80.61	88.16
	Point-Rend	64.42	78.36	98.74	99.37	81.58	88.86
	Fast-SCNN	61.43	76.11	98.60	99.30	80.02	87.70
	DeepLabv3	64.15	78.16	98.73	99.36	81.44	88.76
With short scribble visual prompt	Our PDZSeg	65.45	78.41	98.83	99.41	82.14	88.91
	SETR	70.85	82.94	98.97	99.48	84.91	91.21
	STDC	70.19	82.49	98.94	99.47	84.56	90.98
	Point-Rend	74.73	85.54	99.17	99.58	86.95	92.56
	Fast-SCNN	71.87	83.63	99.04	99.52	85.46	91.58
	DeepLabv3	72.30	83.93	99.08	99.54	85.69	91.73
With long scribble visual prompt	Our PDZSeg	74.06	84.30	99.15	99.56	86.60	91.93