notesum.ai

Published at October 31

Image Synthesis with Class-Aware Semantic Diffusion Models for Surgical Scene Segmentation

cs.CV

cs.AI

Released Date: October 31, 2024

Authors: Yihang Zhou¹, Rebecca Towning², Zaid Awad³, Stamatia Giannarou¹

Aff.: ¹Hamlyn Centre for Robotic Surgery, Department of Surgery and Cancer, Imperial College London, London, SW7 2AZ, UK; ²Imperial College Healthcare NHS Trust, London, W2 1NY, UK; ³Hamlyn Centre for Robotic Surgery, Department of Surgery and Cancer, Imperial College London, London, SW7 2AZ, UK; Imperial College Healthcare NHS Trust, London, W2 1NY, UK

Arxiv: http://arxiv.org/abs/2410.23962v1

Seg. Model	Training Dataset	Blood		Cystic Duct		Hepatic Vein		All
Seg. Model	Training Dataset	mIoU $\uparrow$	mDice $\uparrow$	mIoU $\uparrow$	mDice $\uparrow$	mIoU $\uparrow$	mDice $\uparrow$	mIoU $\uparrow$	mDice $\uparrow$
SegFormer	Original	79.5	88.6	81.8	90.0	65.5	79.1	89.4	94.2
	+ Naive Data Augmentation	78.4	87.9	80.8	89.4	66.1	79.6	89.1	94.0
	+ Conditional DDPM	80.2	89.0	82.5	90.4	66.9	80.2	89.6	94.3
	+ SDM	80.1	88.9	82.7	90.6	66.9	80.2	89.6	94.3
	+ CASDM	80.4	89.1	83.3	90.9	67.2	79.5	89.6	94.3
	+ CASDM + Synthetic Maps	81.5	89.8	83.8	91.2	67.8	80.8	89.9	94.5
Mask2Former	Original	78.9	88.2	80.2	89.0	66.7	80.0	89.3	94.1
	+ Naive Data Augmentation	78.6	88.0	81.2	89.6	65.0	78.8	89.2	94.1
	+ Conditional DDPM	79.2	88.4	81.7	89.9	67.8	80.8	89.7	94.4
	+ SDM	79.2	88.4	81.1	89.6	67.6	80.7	89.6	94.3
	+ CASDM	79.8	88.8	82.2	90.2	67.8	80.8	89.8	94.4
	+ CASDM + Synthetic Maps	80.9	89.4	83.4	90.9	68.5	81.3	90.0	94.5