Image Quality Assessment on CT Reconstruction

Slides:



Advertisements
Similar presentations
An Image Filtering Technique for SPIDER Visible Tomography N. Fonnesu M. Agostini, M. Brombin, R.Pasqualotto, G.Serianni 3rd PhD Event- York- 24th-26th.
Advertisements

Jessica Kishimoto The University of Western Ontario Medical Biophysics Undergrad April 7, 2010 The Effect of Dose on Image Quality When Using an Image.
Image Reconstruction.
QR Code Recognition Based On Image Processing
Object Specific Compressed Sensing by minimizing a weighted L2-norm A. Mahalanobis.
A No-Reference Image Blur Metric Based on the Cumulative Probability of Blur Detection (CPBD) Niranjan D. Narvekar and Lina J. Karam, Senior Member, IEEE.
Image Artifacts Chapter 8 Bushong.
IMAGE QUALITY NOISE LINEARITY CROSS-FIELD UNIFORMITY IMAGE ARTIFACTS.
Guillaume Lavoué Mohamed Chaker Larabi Libor Vasa Université de Lyon
Current Topics in Medical Physics Research Xiaoming Zheng, PhD. School of Dentistry and Health Science Chengdu, China, 2009.
Probabilistic video stabilization using Kalman filtering and mosaicking.
Introduction to Image Quality Assessment
Mean Squared Error : Love It or Leave It ?. Why do we love the MSE ? It is simple. It has a clear physical meaning. The MSE is an excellent metric in.
1 Blind Image Quality Assessment Based on Machine Learning 陈 欣
tomos = slice, graphein = to write
P ERCEPTUAL E VALUATION OF M ULTI -E XPOSURE I MAGE F USION A LGORITHMS Kai Zeng, Kede Ma, Rania Hassen and Zhou Wang Department of Electrical and Computer.
Importance of region-of-interest on image difference metrics Marius Pedersen The Norwegian Color Research Laboratory Faculty of Computer Science and Media.
Information Fusion Yu Cai. Research Article “Comparative Analysis of Some Neural Network Architectures for Data Fusion”, Authors: Juan Cires, PA Romo,
3D CT Image Data Visualize Whole lung tissues Using VTK 8 mm
Frankfurt (Germany), 6-9 June 2011 Xin MIAO, and Xi CHEN – P. R. China – Session 6 – 0393 Communication technical standards infrastructure of the smart.
بسمه تعالی IQA Image Quality Assessment. Introduction Goal : develop quantitative measures that can automatically predict perceived image quality. 1-can.
What is Image Quality Assessment?
Reduction of effective and organ dose to the eye lens in cerebral MDCT scans using iterative image reconstruction Zizka J, Jandura J, Kvasnicka T, Klzo.
1 Iterative Multimodel Subimage Binarization for Handwritten Character Segmentation Author: Amer Dawoud and Mohamed S. Kamel Source: IEEE TRANSACTIONS.
R. Ray and K. Chen, department of Computer Science engineering  Abstract The proposed approach is a distortion-specific blind image quality assessment.
Spatio-Temporal Free-Form Registration of Cardiac MR Image Sequences Antonios Perperidis s /02/2006.
Principles and Practice of Radiation Therapy
Evaluation of Recommender Systems Joonseok Lee Georgia Institute of Technology 2011/04/12 1.
Department of computer science and engineering Evaluation of Two Principal Image Quality Assessment Models Martin Čadík, Pavel Slavík Czech Technical University.
HYPR Project Presentation By Nasser Abbasi HYPR Input-Output view.
Large-Scale Matrix Factorization with Missing Data under Additional Constraints Kaushik Mitra University of Maryland, College Park, MD Sameer Sheoreyy.
J. A. O ’ Sullivan, Quantitative Imaging, 11/19/12 P-20 Seminar, 3/12/05 R. M. Arthur Quantitative Imaging: X-Ray CT and Transmission Tomography Joseph.
Digital Image Processing CSC331 Image restoration 1.
Whiteboard Scanning using Super-resolution STATUS REPORT #2 WODE NI ADVISOR: JOHN MACCORMICK.
Positron Emission Tomography (PET) scans allow for functional imaging of the body’s metabolism, making it an effective tool for locating cancerous tumors.
A 2D/3D correspondence building method for reconstruction of a 3D bone surface model Longwei Fang
1/12 Optimising X-ray computer tomography images with a CT-simulator Philippe Van Marcke K.U.Leuven.
Background Trauma Patients undergo an initial, “on admission” CT scan which includes: Non contrast brain Arterial phase full body scan Portal venous phase.
Blind Image Quality Assessment Using Joint Statistics of Gradient Magnitude and Laplacian Features 刘新 Xue W, Mou X, Zhang L, et al. Blind.
Portland State University Ali Hafiz and Thomas Schumacher
-42 Impact of signal non-repeatability on spectral CT images
Deep Learning for Dual-Energy X-Ray
GPU-based iterative CT reconstruction
You Zhang, Jeffrey Meyer, Joubin Nasehi Tehrani, Jing Wang
Sunday Case of the Day Physics
Evaluation of mA Switching Method with Penalized Weighted Least-Square Noise Reduction for Low-dose CT Yunjeong Lee, Hyekyun Chung, and Seungryong Cho.
Computed Tomography Image Reconstruction
Optimisation of Patient Protection for Radiography
Yuanke Zhang1,2, Hongbing Lu1, Junyan Rong1, Yuxiang Xing3, Jing Meng2
Degradation/Restoration Model
CLASSIFICATION OF TUMOR HISTOPATHOLOGY VIA SPARSE FEATURE LEARNING Nandita M. Nayak1, Hang Chang1, Alexander Borowsky2, Paul Spellman3 and Bahram Parvin1.
Neuro Best Contrast Filter for Head CT
Optimisation of Chest Computed Tomography Using a Phantom: Impact of mAs and Reconstruction Techniques on Image Quality C.S. Reis1; T. Faqir2; V. Harsaker3;
Fig. 1. Images of 43-year-old woman with metastatic lung nodule from rectal cancer show round pulmonary nodule measuring 5 mm (arrows) in left basal lung.
Hayley Milne & Andrew England
Super-resolution Image Reconstruction
Exhibit Number: C19 Evaluation and Reduction of Head Computed Tomography Dose Because of full frame graphic use title slide sparingly because too many.
For Monochromatic Imaging
Enhanced-alignment Measure for Binary Foreground Map Evaluation
Tianfang Li Quantitative Reconstruction for Brain SPECT with Fan-Beam Collimators Nov. 24th, 2003 SPECT system: * Non-uniform attenuation Detector.
Low Dose CT Image Denoising Using WGAN and Perceptual Loss
Pei Qi ECE at UW-Madison
A Review in Quality Measures for Halftoned Images
Fast Hierarchical Back Projection
Prepared by Nuhu Abdulhaqq Isa
Grape Detection in Vineyards
Improving CT quality with optimized image parameters for radiation treatment planning and delivery guidance  Guang-Pei Chen, George Noid, An Tai, Feng.
Image quality measures
Improving Image Accuracy of ROI in CT Using Prior Image
Zhang Xin, Zhao Xiang, Liu Suhong Beijing Normal University
Presentation transcript:

Image Quality Assessment on CT Reconstruction Images: Task-specific vs. General Quality Assessment Jianmei Cai, Xiaogang Chen, Wuhao Huang and Xuanqin Mou Institute of Image Processing and Pattern Recognition Xi’an Jiaotong University June 19, 2017

Introduction Background Experimental methods Results Conclusions

Introduction Tomography Imaging  Medical Diagnosis Enough pathological information Balance between noise and fine image structures Image Quality Assessment (IQA) General IQA task-specific IQA Normal dose Quarter dose The data was provided by the Mayo clinic.

Introduction ALARA[1] (As Low As Reasonably Achievable) Physical layer Algorithm layer  General IQA Diagnosis layer  Task-specific IQA Retrieval layer 1. Physical layer Equipment: Spiral CT, Cone-beam CT Scan protocols: KVP, mAs 2. Algorithm layer Iterative reconstruction: 牵涉到参数选择 Analytic reconstruction: FBP, 牵涉到滤波核的选择? 3. Diagnosis layer Kinds of pathological signals (The influence of the type and shape of the lesion) 4. Retrieval layer [1] A. L. Baert, Encyclopedia of Diagnostic Imaging. Springer Berlin Heidelberg, 2008:60-60.

Background Framework of General IQA Full reference IQA: SSIM, FSIM, GMSD and NLOG-MSE No reference IQA: BIQA 为什么没有RR IQA? General IQA是否要用图像举例,提取适当的纹理

Background Framework of Task-specific IQA Hotelling observer 列比较的公式? Hotelling observer Nonprewhitening observer

Relationship between General and Task-specific IQA Fig.(3) has a overall better visibility than Fig.(2), while the visibility of pathological signal of Fig.(2) is better than Fig.(3). The data was provided by the Mayo clinic.

Relationship between General and Task-specific IQA General IQA and task-specific target on different optimal images The data was provided by the Mayo clinic.

Relationship between General and Task-specific IQA General IQA and task-specific observer target on the same optimal image The data was provided by the Mayo clinic.

Experimental methods Reference images Simulated Lesions: s(r) Background images: b(r) CT images reconstructed by Filter Back Projection (FBP) algorithm 512*512 pixel 𝑠(𝑟)= 𝐶 𝑡 ⋅ 1− 𝑟 𝐷 2 𝜏 示例图片, The CT image background was provided by the Mayo clinic.

Experimental methods Test images  Reference images Add photons Iterative reconstructed Divided into four non-overlapping groups: D1, D2, D3, D4 Subjective experiments The visibility of pathological signal  task-MOS The visibility of the whole image  general-MOS Measurements SRC (Spearman rank order correlation coefficient) PCC (Pearson correlation coefficient) RMSE (Root Mean Squared Error) 打分示例图

Results Performance of general IQA models with general-MOS   D1 D2 D3 D4 srocc pcc rmse SSIM 0.6049 0.6205 0.2898 0.6523 0.7080 0.3766 0.7251 0.7275 0.3659 0.8329 0.8608 0.3058 FSIM 0.5427 0.5455 0.3097 0.4355 0.4786 0.4682 0.8408 0.8351 0.2933 0.7461 0.7765 0.3786 GMSD 0.5217 0.5262 0.3143 0.3092 0.3470 0.5001 0.8510 0.8554 0.2762 0.6476 0.6482 0.4575 NLOG-MSE 0.6464 0.2801 0.5682 0.6095 0.4228 0.7857 0.8002 0.3198 0.7467 0.7685 0.3845 BIQA 0.7071 0.7023 0.2631 0.6256 0.6916 0.3851 0.7166 0.7157 0.3725 0.8471 0.8763 0.2895 Performance of general IQA models with task-MOS BIQA = 0.6458   D1 D2 D3 D4 srocc pcc rmse SSIM 0.0988 0.0980 1.0805 0.1170 0.0836 1.2068 0.0077 0.0010 0.9906 0.0412 0.0615 1.1084 FSIM 0.0729 0.0746 1.0827 0.1221 0.1219 1.2021 0.0342 0.0157 0.9905 0.0416 0.0604 1.1085 GMSD 0.0029 0.0041 1.0857 0.0940 0.0786 1.2073 0.0013 0.0192 0.9904 0.0002 0.0011 1.1105 NLOG-MSE 0.0295 0.0196 1.0855 0.0181 0.0142 1.2110 0.0892 0.1193 0.9835 0.0647 0.0632 1.1083 BIQA 0.2987 0.2093 1.0199 0.3551 0.3769 1.1218 0.3195 0.3419 0.9309 0.2500 0.2831 1.0651

Conclusions The general IQA targets on general quality assessment over the whole image, while task-specific IQA focuses on the visual perception of a specific pathological signal from its background. The relationship between general and task-specific IQA is complex. And the complexity also exists in different general IQA models.

Thank You! Institute of Image Processing and Pattern Recognition Xi’an Jiaotong University June 19, 2017