Cross-sectional relatedness between sentences in breast radiology reports: development of an SVM classifier and evaluation against annotations of five breast radiologists.

Introduce the notion of cross-sectional relatedness as an informational dependence relation between sentences in the conclusion section of a breast radiology report and sentences in the findings section of the same report. Assess inter-rater agreement of breast radiologists. Develop and evaluate a support vector machine (SVM) classifier for automatically detecting cross-sectional relatedness. A standard reference is manually created from 444 breast radiology reports by the first author. A subset of 37 reports is annotated by five breast radiologists. Inter-rater agreement is computed among their annotations and standard reference. Thirteen numerical features are developed to characterize pairs of sentences; the optimal feature set is sought through forward selection. Inter-rater agreement is F-measure 0.623. SVM classifier has F-measure of 0.699 in the 12-fold cross-validation protocol against standard reference. Report length does not correlate with the classifier's performance (correlation coefficient = -0.073). SVM classifier has average F-measure of 0.505 against annotations by breast radiologists. Mediocre inter-rater agreement is possibly caused by: (1) definition is insufficiently actionable, (2) fine-grained nature of cross-sectional relatedness on sentence level, instead of, for instance, on paragraph level, and (3) higher-than-average complexity of 37-report sample. SVM classifier performs better against standard reference than against breast radiologists's annotations. This is supportive of (3). SVM's performance on standard reference is satisfactory. Since optimal feature set is not breast specific, results may transfer to non-breast anatomies. Applications include a smart report viewing environment and data mining.

Robust segmentation of mass-lesions in contrast-enhanced dynamic breast MR images.

PURPOSE: To develop and evaluate a computerized segmentation method for breast MRI (BMRI) mass-lesions. MATERIALS AND METHODS: A computerized segmentation algorithm was developed to segment mass-like-lesions on breast MRI. The segmentation algorithm involved: (i) interactive lesion selection, (ii) automatic intensity threshold estimation, (iii) connected component analysis, and (iv) a postprocessing procedure for hole-filling and leakage removal. Seven observers manually traced the borders of all slices of 30 mass-lesions using the same tools. To initiate the computerized segmentation, each user selected a seed-point for each lesion interactively using two methods: direct seed-point and robust region of interest (ROI) selections. The manual and computerized segmentations were compared pair-wise using the measured size and overlap to evaluate similarity, and the reproducibility of the computerized segmentation was compared with the interobserver variability of the manual delineations. RESULTS: The observed inter- and intraobserver variations were similar (P > 0.05). Computerized segmentation using the robust ROI selection method was significantly (P < 0.001) more reproducible in measuring lesion size (stDev 1.8%) than either manual contouring (11.7%) or computerized segmentation using directly placed seed-point method (13.7%). CONCLUSION: The computerized segmentation method using robust ROI selection is more reproducible than manual delineation in terms of measuring the size of a mass-lesion.

The influence of temporal resolution in determining pharmacokinetic parameters from DCE-MRI data.

We investigated the influence of the temporal resolution of dynamic contrast-enhanced MRI data on pharmacokinetic parameter estimation. Dynamic Gd-DTPA (Gadolinium-diethylene triamine pentaacetic acid) enhanced MRI data of implanted prostate tumors on rat hind limb were acquired at 4.7 T, with a temporal resolution of approximately 5 sec. The data were subsequently downsampled to temporal resolutions in the range of 15 sec to 85 sec, using a strategy that involves a recombination of k-space data. A basic two-compartment model was fit to the contrast agent uptake curves. The results demonstrated that as temporal resolution decreases, the volume transfer constant (K(trans)) is progressively underestimated (approximately 4% to approximately 25%), and the fractional extravascular extracellular space (v(e)) is progressively overestimated (approximately 1% to approximately 10%). The proposed downsampling strategy simulates the influence of temporal resolution more realistically than simply downsampling by removing samples.