notesum.ai

Published at November 19

Enhancing Low Dose Computed Tomography Images Using Consistency Training Techniques

eess.IV

cs.AI

cs.CV

Released Date: November 19, 2024

Authors: Mahmut S. Gokmen¹, Jie Zhang², Ge Wang³, Jin Chen⁴, Cody Bumgardner⁵

Aff.: ¹Department of Computer Science, University of Kentucky; ²Department of Radiology, University of Kentucky; ³Department of Biomedical Engineering, Rensselaer Polytechnic Institute; ⁴Department of Medicine, Department of Biomedical Informatics and Data Science, University of Alabama at Birmingham; ⁵Department of Internal Medicine, Institute for Biomedical Informatics, University of Kentucky

Arxiv: http://arxiv.org/abs/2411.12181v1

Model	#Residual Blocks	NFE	FID $\downarrow$
			CIFAR10 32x32	CelebA 64x64
DDPM [5]	4	1000	3.17	3.26
DDIM [14]	4	10	13.36	17.33
EDM [7]	4	35	2.04	3.32
CT [16]	4	1	14.32	18.72
CT-v1²²2CT-v1: https://github.com/cloneofsimo/consistency_models	4	1	15.54	19.66
iCT [15]	4	1	13.50	15.60
iCT-v1¹¹1iCT-v1: https://github.com/Kinyugo/consistency_models	4	1	14.72	18.54
iCT-v2³³3CT-v2: https://github.com/junhsss/consistency-models	4	1	13.03	16.03
HN-iCT-Small	2	1	10.50	12.31