John J. Heine

MRI segmentation: methods and applications.

The current literature on MRI segmentation methods is reviewed. Particular emphasis is placed on the relative merits of single image versus multispectral segmentation, and supervised versus unsupervised segmentation methods. Image pre-processing and registration are discussed, as well as methods of validation. The application of MRI segmentation for tumor volume measurements during the course of therapy is presented here as an example, illustrating problems associated with inter- and intra-observer variations inherent to supervised methods.

Texture Features from Mammographic Images and Risk of Breast Cancer

Mammographic percent density (PD) is a strong risk factor for breast cancer, but there has been relatively little systematic evaluation of other features in mammographic images that might additionally predict breast cancer risk. We evaluated the association of a large number of image texture features with risk of breast cancer using a clinic-based case-control study of digitized film mammograms, all with screening mammograms before breast cancer diagnosis. The sample was split into training (123 cases and 258 controls) and validation (123 cases and 264 controls) data sets. Age-adjusted and body mass index (BMI)–adjusted odds ratios (OR) per SD change in the feature, 95% confidence intervals, and the area under the receiver operator characteristic curve (AUC) were obtained using logistic regression. A bootstrap approach was used to identify the strongest features in the training data set, and results for features that validated in the second half of the sample were reported using the full data set. The mean age at mammography was 64.0 ± 10.2 years, and the mean time from mammography to breast cancer was 3.7 ± 1.0 (range, 2.0-5.9 years). PD was associated with breast cancer risk (OR, 1.49; 95% confidence interval, 1.25-1.78). The strongest features that validated from each of several classes (Markovian, run length, Laws, wavelet, and Fourier) showed similar ORs as PD and predicted breast cancer at a similar magnitude (AUC = 0.58-0.60) as PD (AUC = 0.58). All of these features were automatically calculated (unlike PD) and measure texture at a coarse scale. These features were moderately correlated with PD (r = 0.39-0.76), and after adjustment for PD, each of the features attenuated only slightly and retained statistical significance. However, simultaneous inclusion of these features in a model with PD did not significantly improve the ability to predict breast cancer. (Cancer Epidemiol Biomarkers Prev 2009;18(3):837–45)

On the statistical nature of mammograms.

We show that digitized mammograms can be considered as evolving from a simple process. A given image results from passing a random input field through a linear filtering operation, where the filter transfer function has a self-similar characteristic. By estimating the functional form of the filter and solving the corresponding filtering equation, the analysis shows that the input field gray value distribution and spectral content can be approximated with parametric methods. The work gives a simple explanation for the variegated image appearance and multimodal character of the gray value distribution common to mammograms. Using the image analysis as a guide, a simulated mammogram is generated that has many statistical characteristics of real mammograms. Additional benefits may follow from understanding the functional form of the filter in conjunction with the input field characteristics that include the approximate parametric description of mammograms, showing the distinction between homogeneously dense and nondense images, and the development of mass analysis methods.

Multiresolution statistical analysis of high-resolution digital mammograms

A multiresolution statistical method for identifying clinically normal tissue in digitized mammograms is used to construct an algorithm for separating normal regions from potentially abnormal regions; that is, small regions that may contain isolated calcifications. This is the initial phase of the development of a general method for the automatic recognition of normal mammograms. The first step is to decompose the image with a wavelet expansion that yields a sum of independent images, each containing different levels of image detail. When calcifications are present, there is strong empirical evidence that only some of the image components are necessary for the purpose of detecting a deviation from normal. The underlying statistic for each of the selected expansion components can be modeled with a simple parametric probability distribution function. This function serves as an instrument for the development of a statistical test that allows for the recognition of normal tissue regions. The distribution function depends on only one parameter, and this parameter itself has an underlying statistical distribution. The values of this parameter define a summary statistic that can be used to set detection error rates. Once the summary statistic is determined, spatial filters that are matched to resolution are applied independently to each selected expansion image. Regions of the image that correlate with the normal statistical model are discarded and regions in disagreement (suspicious areas) are flagged. These results are combined to produce a detection output image consisting only of suspicious areas. This type of detection output is amenable to further processing that may ultimately lead to a fully automated algorithm for the identification of normal mammograms.

An Automated Approach for Estimation of Breast Density

Breast density is a strong risk factor for breast cancer; however, no standard assessment method exists. An automated breast density method was modified and compared with a semi-automated, user-assisted thresholding method (Cumulus method) and the Breast Imaging Reporting and Data System four-category tissue composition measure for their ability to predict future breast cancer risk. The three estimation methods were evaluated in a matched breast cancer case-control (n = 372 and n = 713, respectively) study at the Mayo Clinic using digitized film mammograms. Mammograms from the craniocaudal view of the noncancerous breast were acquired on average 7 years before diagnosis. Two controls with no previous history of breast cancer from the screening practice were matched to each case on age, number of previous screening mammograms, final screening exam date, menopausal status at this date, interval between earliest and latest available mammograms, and residence. Both Pearson linear correlation (R) and Spearman rank correlation (r) coefficients were used for comparing the three methods as appropriate. Conditional logistic regression was used to estimate the risk for breast cancer (odds ratios and 95% confidence intervals) associated with the quartiles of percent breast density (automated breast density method, Cumulus method) or Breast Imaging Reporting and Data System categories. The area under the receiver operator characteristic curve was estimated and used to compare the discriminatory capabilities of each approach. The continuous measures (automated breast density method and Cumulus method) were highly correlated with each other (R = 0.70) but less with Breast Imaging Reporting and Data System (r = 0.49 for automated breast density method and r = 0.57 for Cumulus method). Risk estimates associated with the lowest to highest quartiles of automated breast density method were greater in magnitude [odds ratios: 1.0 (reference), 2.3, 3.0, 5.2; P trend < 0.001] than the corresponding quartiles for the Cumulus method [odds ratios: 1.0 (reference), 1.7, 2.1, and 3.8; P trend < 0.001] and Breast Imaging Reporting and Data System [odds ratios: 1.0 (reference), 1.6, 1.5, 2.6; P trend < 0.001] method. However, all methods similarly discriminated between case and control status; areas under the receiver operator characteristic curve were 0.64, 0.63, and 0.61 for automated breast density method, Cumulus method, and Breast Imaging Reporting and Data System, respectively. The automated breast density method is a viable option for quantitatively assessing breast density from digitized film mammograms. (Cancer Epidemiol Biomarkers Prev 2008;17(11):3090–7)

A Novel Automated Mammographic Density Measure and Breast Cancer Risk

BACKGROUND Mammographic breast density is a strong breast cancer risk factor but is not used in the clinical setting, partly because of a lack of standardization and automation. We developed an automated and objective measurement of the grayscale value variation within a mammogram, evaluated its association with breast cancer, and compared its performance with that of percent density (PD). METHODS Three clinic-based studies were included: a case-cohort study of 217 breast cancer case subjects and 2094 non-case subjects and two case-control studies comprising 928 case subjects and 1039 control subjects and 246 case subjects and 516 control subjects, respectively. Percent density was estimated from digitized mammograms using the computer-assisted Cumulus thresholding program, and variation was estimated from an automated algorithm. We estimated hazards ratios (HRs), odds ratios (ORs), the area under the receiver operating characteristic curve (AUC), and 95% confidence intervals (CIs) using Cox proportional hazards models for the cohort and logistic regression for case-control studies, with adjustment for age and body mass index. We performed a meta-analysis using random study effects to obtain pooled estimates of the associations between the two mammographic measures and breast cancer. All statistical tests were two-sided. RESULTS The variation measure was statistically significantly associated with the risk of breast cancer in all three studies (highest vs lowest quartile: HR = 2.0 [95% CI = 1.3 to 3.1]; OR = 2.7 [95% CI = 2.1 to 3.6]; OR = 2.4 [95% CI = 1.4 to 3.9]; [corrected] all P (trend) < .001). [corrected]. The risk estimates and AUCs for the variation measure were similar to [corrected] those for percent density (AUCs for variation = 0.60-0.62 and [corrected] AUCs for percent density = 0.61-0.65). [corrected]. A meta-analysis of the three studies demonstrated similar associations [corrected] between variation and breast cancer (highest vs lowest quartile: RR = 1.8, 95% CI = 1.4 to 2.3) and [corrected] percent density and breast cancer (highest vs lowest quartile: RR = 2.3, 95% CI = 1.9 to 2.9). CONCLUSION The association between the automated variation measure and the risk of breast cancer is at least as strong as that for percent density. Efforts to further evaluate and translate the variation measure to the clinical setting are warranted.

Review and evaluation of MRI nonuniformity corrections for brain tumor response measurements

Current MRI nonuniformity correction techniques are reviewed and investigated. Many approaches are used to remedy this artifact, but it is not clear which method is the most appropriate in a given situation, as the applications have been with different MRI coils and different clinical applications. In this work four widely used nonuniformity correction techniques are investigated in order to assess the effect on tumor response measurements (change in tumor volume over time): a phantom correction method, an image smoothing technique, homomorphic filtering, and surface fitting approach. Six brain tumor cases with baseline and follow-up MRIs after treatment with varying degrees of difficulty of segmentation were analyzed without and with each of the nonuniformity corrections. Different methods give significantly different correction images, indicating that rf nonuniformity correction is not yet well understood. No improvement in tumor segmentation or in tumor growth/shrinkage assessment was achieved using any of the evaluated corrections.

Spectral analysis of full field digital mammography data.

The spectral content of mammograms acquired from using a full field digital mammography (FFDM) system are analyzed. Fourier methods are used to show that the FFDM image power spectra obey an inverse power law; in an average sense, the images may be considered as 1/f fields. Two data representations are analyzed and compared (1) the raw data, and (2) the logarithm of the raw data. Two methods are employed to analyze the power spectra (1) a technique based on integrating the Fourier plane with octave ring sectioning developed previously, and (2) an approach based on integrating the Fourier plane using rings of constant width developed for this work. Both methods allow theoretical modeling. Numerical analysis indicates that the effects due to the transformation influence the power spectra measurements in a statistically significant manner in the high frequency range. However, this effect has little influence on the inverse power law estimation for a given image regardless of the data representation or the theoretical analysis approach. The analysis is presented from two points of view (1) each image is treated independently with the results presented as distributions, and (2) for a given representation, the entire image collection is treated as an ensemble with the results presented as expected values. In general, the constant ring width analysis forms the foundation for a spectral comparison method for finding spectral differences, from an image distribution sense, after applying a nonlinear transformation to the data. The work also shows that power law estimation may be influenced due to the presence of noise in the higher frequency range, which is consistent with the known attributes of the detector efficiency. The spectral modeling and inverse power law determinations obtained here are in agreement with that obtained from the analysis of digitized film-screen images presented previously. The form of the power spectrum for a given image is approximately l/f2beta with beta approximately 1.4-1.5.

A statistical methodology for mammographic density detection

A statistical methodology is presented based on a chi-square probability analysis that allows the automated discrimination of radiolucent tissue (fat) from radiographic densities (fibroglandular tissue) in digitized mammograms. The method is based on earlier work developed at this facility that shows mammograms may be considered as evolving from a linear filtering operation where a random input field is passed through a 1/f filtering process. The filtering process is reversible which allows the solution of the input field with knowledge obtained from the raw image (the output). The input field solution is analogous to a prewhitening technique or deconvolution. This field contains all the information of the raw image in a much simplified format that can be approximated and analyzed with parametric methods. In the work presented here evidence indicates that there are two random events occurring in the input field with differing variances: (1) one relating to fat tissue with the smaller variance, and (2) the second relating to all other tissue with the larger variance. A statistical comparison of the variances is made by scanning the image with a small search window. A relaxation method allows for making a reliable estimate of the smaller variance which is considered as the global reference. If a local variance deviates significantly from the reference variance, based on chi-square analysis, it is labeled as nonfat; otherwise it is labeled as fat. This statistical test procedure results in a region by region continuous labeling of fat and nonfat tissue across the image. In the work presented here, the emphasis is on the methodology development with supporting preliminary results that are very encouraging. It is widely accepted that mammographic density is a breast cancer risk factor. An important application of this work is to incorporate density-based risk analysis into the ongoing statistical-based detection work developed at this facility. Additional applications include risk analysis dependent on either percentages or total amounts of fat or dense tissue. This work may be considered as the initial step in introducing many of the known breast cancer risk factors into the actual image data analysis.

The influence of mammogram acquisition on the mammographic density and breast cancer association in the mayo mammography health study cohort

IntroductionMammographic density is a strong risk factor for breast cancer. Image acquisition technique varies across mammograms to limit radiation and produce a clinically useful image. We examined whether acquisition technique parameters at the time of mammography were associated with mammographic density and whether the acquisition parameters confounded the density and breast cancer association.MethodsWe examined this question within the Mayo Mammography Health Study (MMHS) cohort, comprised of 19,924 women (51.2% of eligible) seen in the Mayo Clinic mammography screening practice from 2003 to 2006. A case-cohort design, comprising 318 incident breast cancers diagnosed through December 2009 and a random subcohort of 2,259, was used to examine potential confounding of mammogram acquisition technique parameters (x-ray tube voltage peak (kVp), milliampere-seconds (mAs), thickness and compression force) on the density and breast cancer association. The Breast Imaging Reporting and Data System four-category tissue composition measure (BI-RADS) and percent density (PD) (Cumulus program) were estimated from screen-film mammograms at time of enrollment. Spearman correlation coefficients (r) and means (standard deviations) were used to examine the relationship of density measures with acquisition parameters. Hazard ratios (HR) and C-statistics were estimated using Cox proportional hazards regression, adjusting for age, menopausal status, body mass index and postmenopausal hormones. A change in the HR of at least 15% indicated confounding.ResultsAdjusted PD and BI-RADS density were associated with breast cancer (p-trends < 0.001), with a 3 to 4-fold increased risk in the extremely dense vs. fatty BI-RADS categories (HR: 3.0, 95% CI, 1.7 - 5.1) and the ≥ 25% vs. ≤ 5% PD categories (HR: 3.8, 95% CI, 2.5 - 5.9). Of the acquisition parameters, kVp was not correlated with PD (r = 0.04, p = 0.07). Although thickness (r = -0.27, p < 0.001), compression force (r = -0.16, p < 0.001), and mAs (r = -0.06, p = 0.008) were inversely correlated with PD, they did not confound the PD or BI-RADS associations with breast cancer and their inclusion did not improve discriminatory accuracy. Results were similar for associations of dense and non-dense area with breast cancer.ConclusionsWe confirmed a strong association between mammographic density and breast cancer risk that was not confounded by mammogram acquisition technique.