Hendrik Ditt

SIFT and shape context for feature-based nonlinear registration of thoracic CT images

Nonlinear image registration is a prerequisite for various medical image analysis applications. Many data acquisition protocols suffer from problems due to breathing motion which has to be taken into account for further analysis. Intensity based nonlinear registration is often used to align differing images, however this requires a large computational effort, is sensitive to intensity variations and has problems with matching small structures. In this work a feature-based image registration method is proposed that combines runtime efficiency with good registration accuracy by making use of a fully automatic feature matching and registration approach. The algorithm stages are 3D corner detection, calculation of local (SIFT) and global (Shape Context) 3D descriptors, robust feature matching and calculation of a dense displacement field. An evaluation of the algorithm on seven synthetic and four clinical data sets is presented. The quantitative and qualitative evaluations show lower runtime and superior results when compared to the Demons algorithm.

Cervical CT angiography comparing routine noncontrast and a late venous scan as masks for automated bone subtraction: feasibility study and examination of the influence of patient motion on image quality.

Objectives:Bone subtraction techniques have been shown to enhance cranial computed tomography angiography (CTA). The aims of this study were to assess the feasibility of bone subtraction CTA (BSCTA) in cervical CTA, test whether a late venous CT (LVCT) scan can be used as bone mask instead of a low-dose nonenhanced CT (NECT), and to evaluate the impact of patient motion on image quality. Materials and Methods:Thirty-six patients underwent BSCTA for the evaluation of the neck vessels with a 64-slice CT system using commercially available software. Eighteen patients had a low-dose NECT scan before CTA, and 18 patients had an LVCT scan after CTA. Subtraction quality for vascular segments was evaluated independently by 2 examiners. Cohens Kappa was applied to evaluate interobserver reliability, and Wilcoxon signed rank test was used to test for differences between the 2 groups. Motion between the 2 scans was measured and correlated to image quality. Results:BSCTA using both NECT and LVCT scans as masks was successfully applied in all patients. Image quality did not differ significantly between the 2 groups, and interobserver agreement was high (k 0.5–1). Motion between the scans was highest for the jaw and hyoid, and lowest for the upper and lower spine. Decreased image quality on the subtracted images was associated with increased motion for the external carotid and vertebral artery, independent of mask type (P = 0.002–0.04). Conclusions:BSCTA techniques can be successfully applied in the neck. If parenchymal phase imaging is indicated, the LVCT can be used as a bone subtraction mask and diagnostic scan, eg, for tumor imaging.

Whole brain perfused blood volume CT: visualization of infarcted tissue compared to quantitative perfusion CT.

RATIONALE AND OBJECTIVES This study determines the value of whole brain color-coded three-dimensional perfused blood volume (PBV) computed tomography (CT) for the visualization of the infarcted tissue in acute stroke patients. MATERIALS AND METHODS Nonenhanced CT (NECT), perfusion CT (PCT), and CT angiography (CTA) in 48 patients with acute ischemic stroke were performed. Whole brain PBV was calculated from NECT and CTA data sets using commercial software. PBV slices in identical orientation to the PCT slices were reconstructed and the area of visual perfusion abnormality on PBV maps was measured. The infarct core in the corresponding PCT slices (CBV <2.0 mL/100 g) was measured automatically with commercial software. The ischemic area on PBV and the infarct core on quantitative PCT were compared using the Pearsons-R correlation coefficient. Significance was considered for P < .05. RESULTS The quantitative PCT demonstrated a mean infarct core volume of 35.48 +/- 32.17 cm(3), whereas the volume of visual perfusion abnormality of the corresponding PBV slices was 37.16 +/- 37.59 cm(3). The perfusion abnormality in PBV was highly correlated with the infarct core of quantitative PCT for area per slice (r = 0.933, P < .01) as well as volume (r = 0.922, P < .01). CONCLUSIONS PBV can serve as surrogate marker corresponding to the infarct core in acute stroke with whole brain coverage.

Color-coded perfused blood volume imaging using multidetector CT: initial results of whole-brain perfusion analysis in acute cerebral ischemia

Computed tomography (CT) is still the primary imaging modality following acute stroke. To evaluate a prototype of software for the calculation of color-coded whole-brain perfused blood volume (PBV) images from CT angiography (CTA) and nonenhanced CT (NECT) scans, we studied 14 patients with suspected acute ischemia of the anterior cerebral circulation. PBV calculations were performed retrospectively. The detection rate of ischemic changes in the PBV images was compared with NECT. The volume of ischemic changes in PBV was correlated with the infarct volume on follow-up examination taking potential vessel recanalization into account. PBV demonstrated ischemic changes in 12/12 patients with proven infarction and was superior to NECT (8/12) in the detection of early ischemia. Moreover, PBV demonstrated the best correlation coefficient with the follow-up infarct volume (Pearson’s R = 0.957; P = 0.003) for patients with proven recanalization of initially occluded cerebral arteries. In summary, PBV appears to be more accurate in the detection of early infarction compared to NECT and mainly visualizes the irreversibly damaged ischemic tissue.

Bone subtraction CTA for transcranial arteries: intra-individual comparison with standard CTA without bone subtraction and TOF-MRA

AIM To evaluate the impact of bone subtraction computed tomography angiography (BS-CTA) for the assessment of transcranial arteries in comparison with standard CTA (S-CTA) without bone removal and time-of-flight magnetic resonance angiography (TOF-MRA). MATERIALS AND METHODS Cranial unenhanced CT and S-CTA were performed in 53 patients with suspected cerebrovascular disease. BS-CTA datasets were reconstructed from the S-CTA and unenhanced CT source images. TOF-MRA was performed within 24h after CTA on a 1.5 T MRI system. Two radiologists, in consensus, evaluated the segments of the internal carotid artery (C2-C7), the vertebral artery (V4), and the basilar artery for the degree of stenosis. A five-step scale (0-49, 50-69, 70-89, 90-99% and occlusion) for the degree of stenosis was applied for all segments. Wilcoxons signed rank test was used for statistical analysis. RESULTS Seven hundred and fifty vessel segments (ICA:636, VA:106, BA:53) were analysed. The degree of stenosis on S-CTA was consistent with TOF-MRA in all segments. BS-CTA showed a trend towards higher stenosis scores in cases of calcified plaques compared to S-CTA (p=0.11) and TOF-MRA (p=0.09), which was not statistically significant. In transcranial segments, BS-CTA revealed equivalent scores compared to S-CTA and TOF-MRA (p=0.25; p=0.20). CONCLUSION BS-CTA produced similar results to TOF-MRA and S-CTA and can be applied as a non-invasive imaging method for the transcranial arteries. However, BS-CTA shows a trend towards overestimation of the degree of stenosis.

Automatic point landmark matching for regularizing nonlinear intensity registration: application to thoracic CT images

Nonlinear image registration is a prerequisite for a variety of medical image analysis tasks. A frequently used registration method is based on manually or automatically derived point landmarks leading to a sparse displacement field which is densified in a thin-plate spline (TPS) framework. A large problem of TPS interpolation/approximation is the requirement for evenly distributed landmark correspondences over the data set which can rarely be guaranteed by landmark matching algorithms. We propose to overcome this problem by combining the sparse correspondences with intensity-based registration in a generic nonlinear registration scheme based on the calculus of variations. Missing landmark information is compensated by a stronger intensity term, thus combining the strengths of both approaches. An explicit formulation of the generic framework is derived that constrains an intra-modality intensity data term with a regularization term from the corresponding landmarks and an anisotropic image-driven displacement regularization term. An evaluation of this algorithm is performed comparing it to an intensity- and a landmark-based method. Results on four synthetically deformed and four clinical thorax CT data sets at different breathing states are shown.

Improved Follow-Up and Response Monitoring of Thoracic Cage Involvement in Multiple Myeloma Using a Novel CT Postprocessing Software: The Lessons We Learned

OBJECTIVE The purpose of this study is to evaluate the benefit of using novel CT postprocessing software that generates unfolded rib images for more-accurate evaluation of multiple myeloma (MM) at follow-up, response monitoring, and visualization of treatment-related bone changes. MATERIALS AND METHODS Between January 2012 and February 2015, 40 consecutive patients with MM underwent repeated whole-body reduced-dose CT at our institution. The results were retrospectively evaluated and compared with established hematologic markers. Unfolded rib reformatted images were compared with 5- and 1-mm-thick slices with regard to bone changes, bone marrow attenuation, and bone sclerosis. RESULTS Hematologic response categories at follow-up were complete response (CR; n = 2), very good partial response (VGPR; n = 1), partial response (PR; n = 9), stable disease (n = 9), and progressive disease (PD; n = 19). The number of lesions increased in 11 patients (all with PD), decreased in two patients (both with CR), and stayed unchanged in 27 patients. The size of the lesions increased in 14 patients (all with PD), decreased in five patients (two with CR, two with PR, and one with stable disease), and remained unchanged in 21 patients. There was a mean (± SD) difference of 27.99 ± 19.71 HU in bone marrow attenuation for patients with PD (p < 0.0001) and -31.24 ± 13.57 HU in the responders group (p = 0.002), whereas patients with stable disease showed stable bone marrow attenuation at follow-up (mean, -3.37 ± 10.55 HU). Increased bone sclerosis was detected in 12 patients (all of whom were receiving therapy). The sensitivity and specificity of unfolded rib images, 5-mm slices, and 1-mm slices were, respectively, 78.9% and 100%, 52.6% and 100%, and 63.2% and 100% for accurate bone response assessment; 100% and 95.2%, 94.74% and 42.9%, and 89.47% and 47.62% for bone marrow attenuation; and 100% and 100%, 58.3% and 100%, and 91.67% and 100% for sclerosis. CONCLUSION For therapy response assessment, unfolded rib reading is more accurate than transverse CT slices.

Accuracy of Non-Enhanced CT in Detecting Early Ischemic Edema Using Frequency Selective Non-Linear Blending.

Purpose Ischemic brain edema is subtle and hard to detect by computed tomography within the first hours of stroke onset. We hypothesize that non-enhanced CT (NECT) post-processing with frequency-selective non-linear blending (“best contrast”/BC) increases its accuracy in detecting edema and irreversible tissue damage (infarction). Methods We retrospectively analyzed the NECT scans of 76 consecutive patients with ischemic stroke (exclusively middle cerebral artery territory—MCA) before and after post-processing with BC both at baseline before reperfusion therapy and at follow-up (5.73±12.74 days after stroke onset) using the Alberta Stroke Program Early CT Score (ASPECTS). We assessed the differences in ASPECTS between unprocessed and post-processed images and calculated sensitivity, specificity, and predictive values of baseline NECT using follow-up CT serving as reference standard for brain infarction. Results NECT detected brain tissue hypoattenuation in 35 of 76 patients (46.1%). This number increased to 71 patients (93.4%) after post-processing with BC. Follow-up NECT confirmed brain infarctions in 65 patients (85.5%; p = 0.012). Post-processing increased the sensitivity of NECT for brain infarction from 35/65 (54%) to 65/65 (100%), decreased its specificity from 11/11 (100%) to 7/11 (64%), its positive predictive value (PPV) from 35/35 (100%) to 65/69 (94%) and increased its accuracy 46/76 (61%) to 72/76 (95%). Conclusions This post-hoc analysis suggests that post-processing of NECT with BC may increase its sensitivity for ischemic brain damage significantly.

Improved CT Detection of Acute Herpes Simplex Virus Type 1 Encephalitis Based on a Frequency-Selective Nonlinear Blending: Comparison With MRI

OBJECTIVE The purpose of this study is to compare the diagnostic efficacy of a new CT postprocessing tool based on frequency-selective nonlinear blending (best-contrast CT) with that of standard linear blending of unenhanced head CT in patients with herpes simplex virus type 1 and herpes simplex virus encephalitis (HSE), using FLAIR MRI sequences as the standard of reference. MATERIALS AND METHODS Fifteen consecutive patients (six women and nine men; mean [± SD] age, 60 ± 19 years) with proven HSE (positive polymerase chain reaction results from CSF analysis and the presence of neurologic deficits) were retrospectively enrolled. All patients had undergone head CT and MRI (mean time interval, 2 ± 2 days). After standardized unenhanced head CT scans were read, presets of the best-contrast algorithm were determined (center, 30 HU; delta, 5 HU; slope, 5 nondimensional), and resulting images were analyzed. Contrast enhancement was objectively measured by ROI analysis, comparing contrast-to-noise ratios (CNRs) of unenhanced CT and best-contrast CT. FLAIR and DWI MRI sequences were analyzed, and FLAIR was considered as the standard of reference. For assessment of disease extent, a previously reported 50-point score (HSE score) was used. RESULTS CNR values for unenhanced head CT (CNR, 5.42 ± 2.77) could be statistically significantly increased using best-contrast CT (CNR, 9.62 ± 4.28) (p = 0.003). FLAIR sequences yielded a median HSE score of 9.0 (range, 6-17) and DWI sequences yielded HSE scores of 6.0 (range, 5-17). By comparison, unenhanced head CT resulted in a median HSE score of 3.5 (range, 1-6). The median best-contrast CT HSE score was 7.5 (range, 6-10). Agreement between FLAIR and unenhanced CT was 54.44%, that between DWI and best-contrast CT was 95.36%, and that between FLAIR and best-contrast CT was 85.21%. The most frequently overseen findings were located at the level of the upper part of the mesencephalon and at the subthalamic or insular level. CONCLUSION Frequency-selective nonlinear blending significantly increases contrast and detects brain parenchymal involvement in HSE more sensitively compared with unenhanced CT. The sensitivity of best-contrast CT seems to be equal to that of DWI and almost as good as that of FLAIR.

Enhanced reading time efficiency by use of automatically unfolded CT rib reformations in acute trauma.

Rationale of this study was to evaluate whether unfolded rib images enhance time efficiency in detection of rib fractures and time efficiency in patients with acute thoracic trauma. 51 subsequent patients with thoracic trauma underwent 64-slice computed tomography. 1mm thick axial slices were reformatted using a commercially available post-processing software application generating rotatable unfolded rib images. Diagnostic accuracy was evaluated by 3 readers and compared to multiplanar reformations of the original CT images. Reformation and evaluation times were recorded. 116 rib fractures were detected. The multiplanar reformation analysis yielded a sensitivity of 87.9%/93.9%/79.7% with a specificity of 97%/97%/82.2%, whilst the unfolded rib image analysis yielded a sensitivity of 94.8%/94.8%/92.2% and a specificity of 85.2/87.8%/82.4 (p=0.06/0.8/0.04) with high inter-observer agreement (k=0.79-0.85). The mean reading time for the multiplanar reformations was significantly longer (reader 1: 103.7 ± 27.1s/reader 2: 81.8 ± 40.6s/reader 3: 154.3 ± 39.2s) than the evaluation of the unfolded rib images (19.4 ± 4.9s/26.9 ± 15.0s/49.9 ± 18.7s; p<0.01). Concluding, the unfolded rib display reduces reading time for detection of rib fractures in acute thoracic trauma patients significantly and does not compromise the diagnostic accuracy significantly in experienced radiologists. However, unexperienced readers may profit from use of this display.