Karl Sjöstrand

A Novel Automated Platform for Quantifying the Extent of Skeletal Tumour Involvement in Prostate Cancer Patients Using the Bone Scan Index

BACKGROUND There is little consensus on a standard approach to analysing bone scan images. The Bone Scan Index (BSI) is predictive of survival in patients with progressive prostate cancer (PCa), but the popularity of this metric is hampered by the tedium of the manual calculation. OBJECTIVE Develop a fully automated method of quantifying the BSI and determining the clinical value of automated BSI measurements beyond conventional clinical and pathologic features. DESIGN, SETTING, AND PARTICIPANTS We conditioned a computer-assisted diagnosis system identifying metastatic lesions on a bone scan to automatically compute BSI measurements. A training group of 795 bone scans was used in the conditioning process. Independent validation of the method used bone scans obtained ≤3 mo from diagnosis of 384 PCa cases in two large population-based cohorts. An experienced analyser (blinded to case identity, prior BSI, and outcome) scored the BSI measurements twice. We measured prediction of outcome using pretreatment Gleason score, clinical stage, and prostate-specific antigen with models that also incorporated either manual or automated BSI measurements. MEASUREMENTS The agreement between methods was evaluated using Pearsons correlation coefficient. Discrimination between prognostic models was assessed using the concordance index (C-index). RESULTS AND LIMITATIONS Manual and automated BSI measurements were strongly correlated (ρ=0.80), correlated more closely (ρ=0.93) when excluding cases with BSI scores≥10 (1.8%), and were independently associated with PCa death (p<0.0001 for each) when added to the prediction model. Predictive accuracy of the base model (C-index: 0.768; 95% confidence interval [CI], 0.702-0.837) increased to 0.794 (95% CI, 0.727-0.860) by adding manual BSI scoring, and increased to 0.825 (95% CI, 0.754-0.881) by adding automated BSI scoring to the base model. CONCLUSIONS Automated BSI scoring, with its 100% reproducibility, reduces turnaround time, eliminates operator-dependent subjectivity, and provides important clinical information comparable to that of manual BSI scoring.

Sparse modeling of landmark and texture variability using the orthomax criterion

In the past decade, statistical shape modeling has been widely popularized in the medical image analysis community. Predominantly, principal component analysis (PCA) has been employed to model biological shape variability. Here, a reparameterization with orthogonal basis vectors is obtained such that the variance of the input data is maximized. This property drives models toward global shape deformations and has been highly successful in fitting shape models to new images. However, recent literature has indicated that this uncorrelated basis may be suboptimal for exploratory analyses and disease characterization. This paper explores the orthomax class of statistical methods for transforming variable loadings into a simple structure which is more easily interpreted by favoring sparsity. Further, we introduce these transformations into a particular framework traditionally based on PCA; the Active Appearance Models (AAMs). We note that the orthomax transformations are independent of domain dimensionality (2D/3D etc.) and spatial structure. Decompositions of both shape and texture models are carried out. Further, the issue of component ordering is treated by establishing a set of relevant criteria. Experimental results are given on chest radiographs, magnetic resonance images of the brain, and face images. Since pathologies are typically spatially localized, either with respect to shape or texture, we anticipate many medical applications where sparse parameterizations are preferable to the conventional global PCA approach.

Progression of bone metastases in patients with prostate cancer - automated detection of new lesions and calculation of bone scan index

BackgroundThe objective of this study was firstly to develop and evaluate an automated method for the detection of new lesions and changes in bone scan index (BSI) in serial bone scans and secondly to evaluate the prognostic value of the method in a group of patients receiving chemotherapy.MethodsThe automated method for detection of new lesions was evaluated in a group of 266 patients using the classifications by three experienced bone scan readers as a gold standard. The prognostic value of the method was assessed in a group of 31 metastatic hormone-refractory prostate cancer patients who were receiving docetaxel. Cox proportional hazards were used to investigate the association between percentage change in BSI, number of new lesions and overall survival. Kaplan-Meier estimates of the survival function were used to indicate a significant difference between patients with an increase/decrease in BSI or those with two or more new lesions or less than two new lesions.ResultsThe automated method detected progression defined as two or more new lesions with a sensitivity of 93% and a specificity of 87%. In the treatment group, both BSI changes and the number of new metastases were significantly associated with survival. Two-year survival for patients with increasing and decreasing BSI from baseline to follow-up scans were 18% and 57% (p = 0.03), respectively. Two-year survival for patients fulfilling and not fulfilling the criterion of two or more new lesions was 35% and 38% (n.s.), respectively.ConclusionsAn automated method can be used to calculate the number of new lesions and changes in BSI in serial bone scans. These imaging biomarkers contained prognostic information in a small group of patients with prostate cancer receiving chemotherapy.

A path algorithm for the support vector domain description and its application to medical imaging

The support vector domain description is a one-class classification method that estimates the distributional support of a data set. A flexible closed boundary function is used to separate trustworthy data on the inside from outliers on the outside. A single regularization parameter determines the shape of the boundary and the proportion of observations that are regarded as outliers. Picking an appropriate amount of regularization is crucial in most applications but is, for computational reasons, commonly limited to a small collection of parameter values. This paper presents an algorithm where the solutions for all possible values of the regularization parameter are computed at roughly the same computational complexity previously required to obtain a single solution. Such a collection of solutions is known as a regularization path. Knowledge of the entire regularization path not only aids model selection, but may also provide new information about a data set. We illustrate this potential of the method in two applications; one where we establish a sensible ordering among a set of corpora callosa outlines, and one where ischemic segments of the myocardium are detected in patients with acute myocardial infarction.

Robust pseudo-hierarchical support vector clustering

Support vector clustering (SVC) has proven an efficient algorithm for clustering of noisy and high-dimensional data sets, with applications within many fields of research. An inherent problem, however, has been setting the parameters of the SVC algorithm. Using the recent emergence of a method for calculating the entire regularization path of the support vector domain description, we propose a fast method for robust pseudo-hierarchical support vector clustering (HSVC). The method is demonstrated to work well on generated data, as well as for detecting ischemic segments from multidimensional myocardial perfusion magnetic resonance imaging data, giving robust results while drastically reducing the need for parameter estimation.

The entire regularization path for the support vector domain description

The support vector domain description is a one-class classification method that estimates the shape and extent of the distribution of a data set. This separates the data into outliers, outside the decision boundary, and inliers on the inside. The method bears close resemblance to the two-class support vector machine classifier. Recently, it was shown that the regularization path of the support vector machine is piecewise linear, and that the entire path can be computed efficiently. This paper shows that this property carries over to the support vector domain description. Using our results the solution to the one-class classification can be obtained for any amount of regularization with roughly the same computational complexity required to solve for a particularly value of the regularization parameter. The possibility of evaluating the results for any amount of regularization not only offers more accurate and reliable models, but also makes way for new applications. We illustrate the potential of the method by determining the order of inclusion in the model for a set of corpora callosa outlines.

Automated segmentation of the skeleton in whole-body bone scans: influence of difference in atlas

AimAutomated segmentation of the skeleton is the first step for quantitative analysis and computer-aided diagnosis (CAD) of whole-body bone scans. The purpose of this study was to examine the influence of differences in skeletal atlas on the automated segmentation of skeletons in a Japanese patient group. MethodsThe study was based on a bone scan CAD system that included a skeletal atlas obtained using 10 normal bone scans from European patients and 23 normal bone scans from Japanese patients. These were incorporated into the CAD system. The performance of the skeletal segmentation, based on either the European or the Japanese Atlas, was evaluated independently by three observers in a group of 50 randomly selected bone scans from Japanese patients. ResultsThe skeletal segmentation was classified as correct in 41–44 of the 50 cases by the three observers using the Japanese atlas. The corresponding results were 15–18 of the 50 cases using the European atlas, and this difference was statistically significant (P<0.001). The anatomical areas most commonly classified as not correct were the skull, cervical vertebrae, and ribs. ConclusionAutomated segmentation of the skeleton in a Japanese patient group was more successful when the CAD system based on a Japanese atlas was used than when the corresponding system based on a European atlas was used. The results of this study indicate that it is of value to use a skeletal atlas based on normal Japanese bone scans in a CAD system for Japanese patients.

Automatic compartment modelling and segmentation for dynamical renal scintigraphies

Time-resolved medical data has important applications in a large variety of medical applications. In this paper we study automatic analysis of dynamical renal scintigraphies. The traditional analysis pipeline for dynamical renal scintigraphies is to use manual or semiautomatic methods for segmentation of pixels into physical compartments, extract their corresponding time-activity curves and then compute the parameters that are relevant for medical assessment. In this paper we present a fully automatic system that incorporates spatial smoothing constraints, compartment modelling and positivity constraints to produce an interpretation of the full time-resolved data. The method has been tested on renal dynamical scintigraphies with promising results. It is shown that the method indeed produces more compact representations, while keeping the residual of fit low. The parameters of the time activity curve, such as peak-time and time for half activity from peak, are compared between the previous semiautomatic method and the method presented in this paper. It is also shown how to obtain new and clinically relevant features using our novel system.

Area of ischemia assessed by physicians and software packages from myocardial perfusion scintigrams

BackgroundThe European Society of Cardiology recommends that patients with >10% area of ischemia should receive revascularization. We investigated inter-observer variability for the extent of ischemic defects reported by different physicians and by different software tools, and if inter-observer variability was reduced when the physicians were provided with a computerized suggestion of the defects.MethodsTwenty-five myocardial perfusion single photon emission computed tomography (SPECT) patients who were regarded as ischemic according to the final report were included. Eleven physicians in nuclear medicine delineated the extent of the ischemic defects. After at least two weeks, they delineated the defects again, and were this time provided a suggestion of the defect delineation by EXINI HeartTM (EXINI). Summed difference scores and ischemic extent values were obtained from four software programs.ResultsThe median extent values obtained from the 11 physicians varied between 8% and 34%, and between 9% and 16% for the software programs. For all 25 patients, mean extent obtained from EXINI was 17.0% (± standard deviation (SD) 14.6%). Mean extent for physicians was 22.6% (± 15.6%) for the first delineation and 19.1% (± 14.9%) for the evaluation where they were provided computerized suggestion. Intra-class correlation (ICC) increased from 0.56 (95% confidence interval (CI) 0.41-0.72) to 0.81 (95% CI 0.71-0.90) between the first and the second delineation, and SD between physicians were 7.8 (first) and 5.9 (second delineation).ConclusionsThere was large variability in the estimated ischemic defect size obtained both from different physicians and from different software packages. When the physicians were provided with a suggested delineation, the inter-observer variability decreased significantly.

Sparse statistical deformation model for the analysis of craniofacial malformations in the Crouzon mouse

Crouzon syndrome is characterised by the premature fusion of cranial sutures. Recently the first genetic Crouzon mouse model was generated. In this study, Micro CT skull scannings of wild-type mice and Crouzon mice were investigated. Using nonrigid registration, a wild-type craniofacial mouse atlas was built. The atlas was registered to all mice providing parameters controlling the deformations for each subject. Our previous PCA-based statistical deformation model on these parameters revealed only one discriminating mode of variation. Aiming at distributing the discriminating variation over more modes we built a different model using Independent Component Analysis (ICA). Here, we focus on a third method, sparse PCA (SPCA), which aims at approximating the properties of a standard PCA while introducing sparse modes of variation. The results show that SPCA outperforms both ICA and PCA with respect to the Fisher discriminant, although many similarities are found with respect to ICA.