Classification of small lesions on dynamic breast MRI: Integrating dimension reduction and out-of-sample extension into CADx methodology.

OBJECTIVE: While dimension reduction has been previously explored in computer aided diagnosis (CADx) as an alternative to feature selection, previous implementations of its integration into CADx do not ensure strict separation between training and test data required for the machine learning task. This compromises the integrity of the independent test set, which serves as the basis for evaluating classifier performance. METHODS AND MATERIALS: We propose, implement and evaluate an improved CADx methodology where strict separation is maintained. This is achieved by subjecting the training data alone to dimension reduction; the test data is subsequently processed with out-of-sample extension methods. Our approach is demonstrated in the research context of classifying small diagnostically challenging lesions annotated on dynamic breast magnetic resonance imaging (MRI) studies. The lesions were dynamically characterized through topological feature vectors derived from Minkowski functionals. These feature vectors were then subject to dimension reduction with different linear and non-linear algorithms applied in conjunction with out-of-sample extension techniques. This was followed by classification through supervised learning with support vector regression. Area under the receiver-operating characteristic curve (AUC) was evaluated as the metric of classifier performance. RESULTS: Of the feature vectors investigated, the best performance was observed with Minkowski functional 'perimeter' while comparable performance was observed with 'area'. Of the dimension reduction algorithms tested with 'perimeter', the best performance was observed with Sammon's mapping (0.84±0.10) while comparable performance was achieved with exploratory observation machine (0.82±0.09) and principal component analysis (0.80±0.10). CONCLUSIONS: The results reported in this study with the proposed CADx methodology present a significant improvement over previous results reported with such small lesions on dynamic breast MRI. In particular, non-linear algorithms for dimension reduction exhibited better classification performance than linear approaches, when integrated into our CADx methodology. We also note that while dimension reduction techniques may not necessarily provide an improvement in classification performance over feature selection, they do allow for a higher degree of feature compaction.

Classification of Small Lesions in Breast MRI: Evaluating The Role of Dynamically Extracted Texture Features Through Feature Selection.

Dynamic texture quantification, i.e., extracting texture features from the lesion enhancement pattern in all available post-contrast images, has not been evaluated in terms of its ability to classify small lesions. This study investigates the classification performance achieved with texture features extracted from all five post-contrast images of lesions (mean lesion diameter of 1.1 cm) annotated in dynamic breast magnetic resonance imaging exams. Sixty lesions are characterized dynamically using Haralick texture features. The texture features are then used in a classification task with support vector regression and a fuzzy k-nearest neighbor classifier; free parameters of these classifiers are optimized using random sub-sampling cross-validation. Classifier performance is determined through receiver-operator characteristic (ROC) analysis, specifically through computation of the area under the ROC curve (AUC). Mutual information is used to evaluate the contribution of texture features extracted from different post-contrast stages to classifier performance. Significant improvements (p < 0.05) are observed for six of the thirteen texture features when the lesion enhancement pattern is quantified using the proposed approach of dynamic texture quantification. The highest AUC value observed (0.82) is achieved with texture features responsible for capturing aspects of lesion heterogeneity. Mutual information analysis reveals that texture features extracted from the third and fourth post-contrast images contributed most to the observed improvement in classifier performance. These results show that the performance of automated character classification with small lesions can be significantly improved through dynamic texture quantification of the lesion enhancement pattern.

Classification of small lesions in dynamic breast MRI: Eliminating the need for precise lesion segmentation through spatio-temporal analysis of contrast enhancement over time.

Characterizing the dignity of breast lesions as benign or malignant is specifically difficult for small lesions; they don't exhibit typical characteristics of malignancy and are harder to segment since margins are harder to visualize. Previous attempts at using dynamic or morphologic criteria to classify small lesions (mean lesion diameter of about 1 cm) have not yielded satisfactory results. The goal of this work was to improve the classification performance in such small diagnostically challenging lesions while concurrently eliminating the need for precise lesion segmentation. To this end, we introduce a method for topological characterization of lesion enhancement patterns over time. Three Minkowski Functionals were extracted from all five post-contrast images of sixty annotated lesions on dynamic breast MRI exams. For each Minkowski Functional, topological features extracted from each post-contrast image of the lesions were combined into a high-dimensional texture feature vector. These feature vectors were classified in a machine learning task with support vector regression. For comparison, conventional Haralick texture features derived from gray-level co-occurrence matrices (GLCM) were also used. A new method for extracting thresholded GLCM features was also introduced and investigated here. The best classification performance was observed with Minkowski Functionals area and perimeter, thresholded GLCM features f8 and f9, and conventional GLCM features f4 and f6. However, both Minkowski Functionals and thresholded GLCM achieved such results without lesion segmentation while the performance of GLCM features significantly deteriorated when lesions were not segmented (p < 0.05). This suggests that such advanced spatio-temporal characterization can improve the classification performance achieved in such small lesions, while simultaneously eliminating the need for precise segmentation.

Numerical comparison of X-ray differential phase contrast and attenuation contrast.

We present a numerical tool to compare directly the contrast-to-noise-ratio (CNR) of the attenuation- and differential phase-contrast signals available from grating-based X-ray imaging for single radiographs. The attenuation projection is differentiated to bring it into a modality comparable to the differential phase projection using a Gaussian derivative filter. A Relative Contrast Gain (RCG) is then defined as the ratio of the CNR of image values in a region of interest (ROI) in the differential phase projection to the CNR of image values in the same ROI in the differential attenuation projection. We apply the method on experimental data of human breast tissue acquired using a grating interferometer to compare the two contrast modes for two regions of interest differing in the type of tissue. Our results indicate that the proposed method can be used as a local estimate of the spatial distribution of the ratio δ/ß, i.e., real and imaginary part of the complex refractive index, across a sample.

Dynamic breast MRI in the course of neoadjuvant chemotherapy: standardized evaluation of tumor size and enhancement parameters in correlation to different histopathologic characteristics.

RATIONALE AND OBJECTIVES: Basic exploratory data analysis to evaluate enhancement and tumor size (SIZE) in contrast-enhanced breast magnetic resonance imaging (CE-MRI) during chemotherapy. Correlation with histopathology (human epidermal growth factor receptor (HER2/neu) status and estrogen receptor (ER) score). MATERIALS AND METHODS: Sixty-five women (mean age 47 +/- 10 years) with locally advanced breast cancer (mean SIZE 25 mL) had CE-MRI (three-dimensional fast low angle shot (FLASH); repetition time = 9.1 ms, echo time = 4.8 ms, flip angle (FA) 25 degrees, matrix size 256 x 256 pixels, field of view 350 mm, slice thickness 2 mm, number of slices = 32, one precontrast and five postcontrast series) before and after chemotherapy. Lesion segmentation and subsequent SIZE and enhancement analysis including maximum enhancement (MAX), area under the curve (AUC), time-to-peak (TTP), and maximum upslope (MUS) were performed. Correlation with histopathology (ER score and HER2/neu status). RESULTS: SIZE reduced significantly during therapy (25 mL vs. 5 mL, P < .0001). AUC, MAX, MUS decreased (P < .0001), TTP increased (P < .0001). SIZE and MAX were independent parameters (r(2) = .22). No correlation (P > .01) in any of the parameters with either ER score or HER2/neu status was found. HER2/neu score equal 2+pos. or 3+ showed significantly stronger changes in SIZE (P < .01), MAX (P < .01) and AUC (P < .05) compared to lower HER2/neu score (0 to 2+neg.). CONCLUSIONS: From routine MRI protocol and semiquantitative analysis of signal enhancement curves, information about size, and hemodynamic status of tumors under treatment may be extracted. Reduction in size and maximum enhancement were complementary parameters. In the course of therapy, size and enhancement may develop differently in clinically relevant histopathological subgroups.

Automated classification of breast parenchymal density: topologic analysis of x-ray attenuation patterns depicted with digital mammography.

OBJECTIVE: We used an algorithm for quantitative image processing to classify breast tissue into the categories fibrosis, involution atrophy, and normal. The algorithm entailed use of Minkowski functionals in topologic analysis of x-ray attenuation patterns on digital mammograms. The results were compared with those of techniques based on evaluation of gray-level histograms. MATERIALS AND METHODS: One hundred digital mammograms were classified by consensus of two experienced readers. A topologic parameter extracted from the Minkowski functional spectra was obtained for retromammilar image sections (512 x 512 pixels). From the gray-level histogram of each of these samples, the 20th percentile, median, and mean were determined. Discriminant analysis was used to assess the predictive value of the methods with respect to correct categorization. RESULTS: The mean gray-level intensity of normal breast tissue was 90 +/- 9, and the 20th percentile was 68 +/- 18. The mean gray-level intensity was 84 +/- 7 for involution and 90 +/- 8 for fibrosis; the 20th percentile was 75 +/- 6 for involution and 73 +/- 10 for fibrosis. The results of discriminant analysis showed that use of the gray-level histogram parameters led to correct classification in 66% of cases. Use of topologic analysis with Minkowski functionals increased the rate of correct classification to 83%. When a combined model of histogram-derived parameters and Minkowski functionals was used, 89% of cases were categorized correctly. CONCLUSION: Topologic analysis of x-ray attenuation patterns on digital mammograms obtained with Minkowski functionals is simple and robust, and the results agree with radiologists' ratings. Because correct classification is significantly higher than with use of density features, our technique may be an objective and quantitative alternative in the evaluation of the parenchymal structure of the breast.

Classification of small contrast enhancing breast lesions in dynamic magnetic resonance imaging using a combination of morphological criteria and dynamic analysis based on unsupervised vector-quantization.

PURPOSE: To evaluate the diagnostic value of breast magnetic resonance imaging (MRI) in small focal lesions using dynamic analysis based on unsupervised vector quantization in combination with a score for morphologic criteria. MATERIALS AND METHODS: We examined 85 mammographically indetermintate lesions (BIRADS 3-4; 47 malignant, mean lesion size 1.2 cm; 38 benign, mean lesion size 1.1 cm). MRI was performed with a dynamic T1-weighted gradient echo sequence (1 precontrast and 5 postcontrast series). Lesions with an initial contrast enhancement >/=50% were selected with semiautomatic segmentation. For conventional dynamic analysis, we calculated the mean initial signal increase and postinitial course of all voxels included in a lesion. Secondly, all voxels within the lesions were assigned to 4 clusters using minimal-free-energy vector quantization. Dynamic and morphologic criteria were summarized in a diagnostic score and evaluated by receiver operating characteristic analysis. RESULTS: In the present collection of small lesions, morphologic criteria [area under the curve (AUC) = 0.610] were inferior to dynamic criteria in the detection of breast cancer. Dynamic analysis with vector quantization (AUC = 0.760) presented slightly better results compared with standard dynamic analysis (AUC = 0.693). There was no benefit for combined morphologic and dynamic analysis. CONCLUSION: In small MR-mammographic lesions, dynamic analysis with vector quantization alone tends to result in a higher diagnostic accuracy compared with combined morphologic and dynamic analysis.

Cluster analysis of signal-intensity time course in dynamic breast MRI: does unsupervised vector quantization help to evaluate small mammographic lesions?

We examined whether neural network clustering could support the characterization of diagnostically challenging breast lesions in dynamic magnetic resonance imaging (MRI). We examined 88 patients with 92 breast lesions (51 malignant, 41 benign). Lesions were detected by mammography and classified Breast Imaging and Reporting Data System (BIRADS) III (median diameter 14 mm). MRI was performed with a dynamic T1-weighted gradient echo sequence (one precontrast and five postcontrast series). Lesions with an initial contrast enhancement >or=50% were selected with semiautomatic segmentation. For conventional analysis, we calculated the mean initial signal increase and postinitial course of all voxels included in a lesion. Secondly, all voxels within the lesions were divided into four clusters using minimal-free-energy vector quantization (VQ). With conventional analysis, maximum accuracy in detecting breast cancer was 71%. With VQ, a maximum accuracy of 75% was observed. The slight improvement using VQ was mainly achieved by an increase of sensitivity, especially in invasive lobular carcinoma and ductal carcinoma in situ (DCIS). For lesion size, a high correlation between different observers was found (R(2) = 0.98). VQ slightly improved the discrimination between malignant and benign indeterminate lesions (BIRADS III) in comparison with a standard evaluation method.