notesum.ai

Published at November 27

Neural Finite-State Machines for Surgical Phase Recognition

eess.IV

cs.CV

Released Date: November 27, 2024

Authors: Hao Ding¹, Zhongpai Gao², Benjamin Planche², Tianyu Luan³, Abhishek Sharma², Meng Zheng², Ange Lou⁴, Terrence Chen², Mathias Unberath¹, Ziyan Wu²

Aff.: ¹Johns Hopkins University, Baltimore, MD, USA; ²United Imaging Intelligence, Boston MA, USA; ³State University of New York at Buffalo, Buffalo, NY, USA; ⁴Vanderbilt University, Nashville, TN, USA

Arxiv: http://arxiv.org/abs/2411.18018v1

Datasets	Methods	Relaxed Metric	Video-level Accuracy	Phase-level
Datasets	Methods	Relaxed Metric	Video-level Accuracy	Precision	Recall	Jaccard
	SV-RCNet [17]	✓	$85.3\pm 7.3$	$80.7$	$83.5$	$-$
	TMRNet [19]	✓	$90.1\pm 7.6$	$90.3$	$89.5$	$79.1$
	TeCNO [7]	✓	$88.6\pm 7.8$	$86.5$	$87.6$	$75.1$
	Trans-SVNet [14]	✓	$90.3\pm 7.1$	$90.7$	$88.8$	$79.3$
	Not End-to-End [39]	✓	$91.5\pm 7.1$	$-$	$86.8$	$77.2$
	LoViT [26]	✓	$92.4\pm 6.3$	$89.9$	$90.6$	$81.2$
	SKiT [27]	✓	$\pagecolor{lightbrown}93.4\pm 5.2$	$90.9$	$91.8$	$82.6$
\cdashline2-7	Encoder-Decoder	✓	$89.6\pm 7.6$	$87.5$	$88.0$	$76.6$
	Encoder-Decoder + NFSM (Ours)	✓	$92.3\pm 5.0$ _(+2.7)	$90.4$ _(+2.9)	$90.5$ _(+2.5)	80.9 _(+3.3)
	Surgformer [37]	✓	$93.3\pm 6.2$	$91.8$	$92.3$	$84.3$
Cholec80 [32]	Surgformer + NFSM (Ours)	✓	93.7 $\pm$ 5.5 _(+0.4)	93.0 _(+1.2)	92.7 _(+0.4)	85.2 _(+0.9)
	AVT [15]		$86.7\pm 7.0$	$77.3$	$82.1$	$66.4$
	TeSTra [40]		$90.1\pm 7.6$	$82.8$	$83.8$	$71.6$
	Trans-SVNet [14]		$89.1\pm 6.6$	$84.7$	$83.6$	$72.5$
	LoViT [26]		$91.5\pm 6.1$	$83.1$	$86.5$	$74.2$
	SKiT [27]		92.5 $\pm$ 5.1	$84.6$	88.5	$76.7$
\cdashline2-7	Encoder-Decoder		$88.8\pm 7.5$	$82.6$	$83.0$	$70.8$
	Encoder-Decoder + NFSM (Ours)		$91.1\pm 5.1$ _(+2.3)	$85.8$ _(+3.2)	$86.0$ _(+3.0)	75.6 _(+4.8)
	Surgformer [37]		$92.3\pm 6.2$	$87.1$	$87.6$	$77.8$
	Surgformer + NFSM (Ours)		$92.3\pm 5.4$ _(+0.0)	87.3 _(+0.2)	$87.7$ _(+0.1)	78.0 _(+0.2)
	SV-RCNet [17]		$75.6$	$64.0$	$59.7$	$47.2$
	TMRNet [19]		$78.2$	$66.0$	$61.5$	$49.6$
	TeCNO [7]		$77.3$	$66.9$	$64.6$	$50.7$
	AVT [15]		$77.8$	$64.2$	$62.1$	$50.7$
AutoLaparo [36]	Trans-SVNet [14]		$78.3$	$68.0$	$62.2$	$50.7$
	LoViT [26]		$81.4\pm 7.6$	85.1	$65.9$	$55.9$
	SKiT [27]		$82.9\pm 6.8$	81.8	$70.1$	$59.9$
\cdashline2-7	Encoder-Decoder		$81.8\pm 7.3$	$76.5$	$66.2$	$56.2$
	Encoder-Decoder + NFSM (Ours)		$84.0\pm 7.2$ _(+2.2)	$74.9$	$66.8$ _(+0.6)	58.1 _(+1.9)
	Surgformer [37]		$86.1\pm 7.3$	81.5	70.8	62.4
	Surgformer + NFSM (Ours)		86.7 $\pm$ 7.2 _(+0.6)	84.8 _(+3.3)	71.1 _(+0.3)	63.2 _(+0.8)