1、從腦波會溝通 談醫學影像處,1,李宗南國立中山大學 資訊工程系March 1, 2010,2,3,Medical Imaging,Medical imaging is one of the most commonly used methods, other than human observation and biochemical exam., used in modern clinical diagnoses and treatments Medical imaging industry has been the most substantial and profitable part i
2、n medical industry Decrease in imaging cost and increase in imaging capability makes the applications more popular in clinics,4,Medical Imaging,Provide two major types of imaging capabilities: anatomical and functional imaging Involve signal processing and image processing to achieve better image qu
3、ality and cost Integrate computer vision processing, graphics, and virtual reality,5,Image Modality,Based on Interested Physical Property: X-Ray (CT/Radiography) MRI (Magnetic Resonance Imaging) PET (Positron Emission Tomography) US (Ultra Sound) SPECT (Single Photon Emission CT) EIT (Electrical Imp
4、edance Tomography) Video and etc.,6,Image acquisition,Registration,Segmentation,3D reconstruction,Visualization,Data analysis,Medical Image Analysis Flow,7,Segmentation,8,Segmentation,9,Visualization,10,Visualization,11,Some researches,Robust Motion Correction for Functional MR Imaging of the BrainA
5、utomated Nasopharyngeal Carcinoma DetectionProjection Profile Analysis for Identifying Different Views of Chest Radiographs,12,Robust Motion Correction for Functional MR Imaging of the Brain許正忠 吳銘庭 李宗南,13,A Volume Functional MR Images,14,Motor task paradigmThe performance of the algorithms have been
6、 visually compared to the AIR 2.0 and SPM 96 packages with the color-coded statistical mapping by Kolmogorov-Smirov method.,Registration of fMRI,15,Registration of fMRI (cont.),The color-coded statistical maps by the Kolmogorov-Smirov method,Original,Tukeys biweight,Skipped mean,Red-yellow: positive
7、 correlation; blue-green: negative correlation,16,Registration of fMRI (cont.),The color-coded statistical maps by the Kolmogorov-Smirov method,AIR 2.0,SPM 96,17,Registration of fMRI (cont.),The time-signal intensity course from the region of interest that consists of four voxels,Upper: Original; Tu
8、keys biweight; Skipped mean Bottom: AIR 2.0; SPM 96,18,Registration of Dynamic MRI,Upper: The slices at 0 and 180 seconds before motion correction Bottom-left: Difference between the image at 0 second and the image at 180 seconds before motion correction Bottom-right: Difference between the register
9、ed image and the image in upper-left,19,Automated Nasopharyngeal Carcinoma Detection with Dynamic Gadolinium-Enhanced MRI許正忠 賴炳宏 李宗南 黃文禎,20,Introduction,We have applied dynamic gadolinium-enhanced MR imaging to distinguishing recurrent nasal papilloma from postoperative changes and had satisfying re
10、sults. Hence, we evaluate the feasibility of dynamic gadolinium-enhanced MR imaging in the detection of nasopharyngeal carcinoma (NPC).,21,Introduction (cont.),In a prior work, finding the ROI was done in a semi-automatic manner.,22,Methods,Registration of Dynamic Gadolinium-Enhanced MR Images Head
11、Mask Generation Dynamic Gadolinium-Enhanced MR Data Quantitative Evaluation The Relative Signal Increase Method The Slope Method The Relative Signal Change Method,23,Methods (cont.),Rough Segmentation Rough Segmentation Refinement Morphological operations A binary opening operation was performed fir
12、st on the binary image to remove noise. A binary closing operation was then performed on the opened image to display the necrosis region that is a part of the NPC. Region labeling and the largest region was chosen as the NPC,24,Head Mask Generation,left: original image; right: head mask,25,The Relat
13、ive Signal Increase (RSI) Method,SIcm (t)/SI0 RSI (t) = 1 + A 1 - exp(-t /Tc) - Ct,We used A and Tc to characterize tissue. Note that the calculated A is 2.1 and Tc is 17.3 for NPC.,26,Two New Approach,However, the nonlinear least-squares fitting process in the RSI method is time-consuming. We prese
14、nt two faster approaches to compute the relative intensity difference between dynamic MR images.,27,The Slope Method,Slope(ti) = (SIcm (ti) - SIcm (ti-1) / (ti - ti-1) MS(ti) = Max(Slope(t1), Slope(t2), , Slope(tn),We used the calculated MS(ti) and ti to characterize tissue. Note that the calculated
15、 MS is 2.18, occurring at 30 seconds for NPC.,28,The Relative Signal Change (RSC) Method,RSC(ti) = SIcm(ti) / SIcm(ti-1) MRSC(ti) = Max(RSC(t1), RSC(t2), , RSC(tn),We used the calculated MRSC(ti) and ti to characterize tissue. Note that the calculated MRSC is 2.47, occurring at 30 seconds for NPC.,2
16、9,The Criteria for Rough Segmentation,For the Relative Signal Increase method,A candidate for NPC can be defined as the combination of enhancement amplitude = 0.8 (arbitrary units) and distribution time =45 (seconds).,30,The Criteria for Rough Segmentation (cont.),For the Slope method,A candidate fo
17、r NPC can be defined as the combination of maximal slope = 1.1 (arbitrary units) and time =30 (seconds).,31,The Criteria for Rough Segmentation (cont.),For the Relative Signal Change method,A candidate for NPC can be defined as the combination of maximal relative signal change = 1.5 (arbitrary units
18、) and time =30 (seconds).,32,Rough Segmentation,left: RSI; middle: Slope; right: RSC,33,Rough Segmentation Refinement,:,Upper-left: RSI; Upper-middle: Slope Upper-right: RSC; Bottom: radiologist hand-labeled,34,Rough Segmentation Refinement (cont.),left: RSI; middle: Slope; right: RSC,35,Compared to
19、 Radiologist Hand-Labeled Segmentation,left: RSI; middle: Slope; right: RSC,Red: true positive Green: false positive Blue: false negative,36,Dynamic Gadolinium-Enhanced MR Imaging,37,Projection Profile Analysis for Identifying Different Views of Chest RadiographsE-Fong Kao, PhD, Chung-Nan Lee, PhD, Twei-Shiun Jaw, MD, Jui-Sheng Hsu, MD, PhD, Gin-Chung Liu, MD 趙垂勳 許瑞昇 劉金昌,38,Overall scheme for the projection profile analysis in chest images.,39,40,
