Biomedical subjects

Joseph Y Lo

Publications and source records attributed to Joseph Y Lo.

17 recordsLinked to original sources

PubMed2006

Introduction to neutron stimulated emission computed tomography.

Neutron stimulated emission computed tomography (NSECT) is presented as a new technique for in vivo tomographic spectroscopic imaging. A full implementation of NSECT is intended to provide an elemental spectrum of the body or part of the body being interrogated at each voxel of a three-dimensional computed tomographic image. An external neutron beam illuminates the sample and some of these neutrons scatter inelastically, producing characteristic gamma emission from the scattering nuclei. These characteristic gamma rays are acquired by a gamma spectrometer and the emitting nucleus is identified by the emitted gamma energy. The neutron beam is scanned over the body in a geometry that allows for tomographic reconstruction. Tomographic images of each element in the spectrum can be reconstructed to represent the spatial distribution of elements within the sample. Here we offer proof of concept for the NSECT method, present the first single projection spectra acquired from multi-element phantoms, and discuss potential biomedical applications.

Gamma Rays↗

PubMed2006

Optimized approach to decision fusion of heterogeneous data for breast cancer diagnosis.

As more diagnostic testing options become available to physicians, it becomes more difficult to combine various types of medical information together in order to optimize the overall diagnosis. To improve diagnostic performance, here we introduce an approach to optimize a decision-fusion technique to combine heterogeneous information, such as from different modalities, feature categories, or institutions. For classifier comparison we used two performance metrics: The receiving operator characteristic (ROC) area under the curve [area under the ROC curve (AUC)] and the normalized partial area under the curve (pAUC). This study used four classifiers: Linear discriminant analysis (LDA), artificial neural network (ANN), and two variants of our decision-fusion technique, AUC-optimized (DF-A) and pAUC-optimized (DF-P) decision fusion. We applied each of these classifiers with 100-fold cross-validation to two heterogeneous breast cancer data sets: One of mass lesion features and a much more challenging one of microcalcification lesion features. For the calcification data set, DF-A outperformed the other classifiers in terms of AUC (p < 0.02) and achieved AUC=0.85 +/- 0.01. The DF-P surpassed the other classifiers in terms of pAUC (p < 0.01) and reached pAUC=0.38 +/- 0.02. For the mass data set, DF-A outperformed both the ANN and the LDA (p < 0.04) and achieved AUC=0.94 +/- 0.01. Although for this data set there were no statistically significant differences among the classifiers' pAUC values (pAUC=0.57 +/- 0.07 to 0.67 +/- 0.05, p > 0.10), the DF-P did significantly improve specificity versus the LDA at both 98% and 100% sensitivity (p < 0.04). In conclusion, decision fusion directly optimized clinically significant performance measures, such as AUC and pAUC, and sometimes outperformed two well-known machine-learning techniques when applied to two different breast cancer data sets.

Algorithms↗

PubMed2005

Comparative scatter and dose performance of slot-scan and full-field digital chest radiography systems.

PURPOSE: To evaluate the scatter, dose, and effective detective quantum efficiency (DQE) performance of a slot-scan digital chest radiography system compared with that of a full-field digital radiography system. MATERIALS AND METHODS: Scatter fraction of a slot-scan system was measured for an anthropomorphic and a geometric phantom by using a posterior beam-stop technique at 117 and 140 kVp. Measurements were repeated with a full-field digital radiography system with and without a 13:1 antiscatter grid at 120 and 140 kVp. For both systems, the effective dose was measured on posteroanterior and lateral views for standard clinical techniques by using dosimeters embedded in a female phantom. The effective DQEs of the two systems were assessed by taking into account the scatter performance and the DQE of each system. The statistical significance of all the comparative differences was ascertained by means of t test analysis. RESULTS: The slot-scan system and the full-field system with grid yielded scatter fractions of 0.13-0.14 and 0.42-0.48 in the lungs and 0.30-0.43 and 0.69-0.78 in the mediastinum, respectively. The sum of the effective doses for posteroanterior and lateral views for the slot-scan system (0.057 mSv +/- 0.003 [+/- standard deviation]) was 34% lower than that for the full-field system (0.086 mSv +/- 0.001, P < .05) at their respective clinical peak voltages (140 and 120 kVp, respectively). The effective DQE of the slot-scan system was equivalent to that of the full-field system in the lung region but was 37% higher in the dense regions (P < .05). CONCLUSION: The slot-scan design leads to marked scatter reduction compared with the more conventional full-field geometries with a grid. The improved scatter performance of a slot-scan geometry can effectively compensate for low DQE and lead to improved image quality.

Phantoms, Imaging↗

PubMed2005

Accuracy of segmentation of a commercial computer-aided detection system for mammography.

PURPOSE: To assess the accuracy of segmentation in a commercially available computer-aided detection (CAD) system. MATERIALS AND METHODS: Approval for this study was obtained from the authors' institutional review board. Informed consent was not required by the board for this review, as data were stripped of patient identifiers. Two thousand twenty mammograms from 507 women were analyzed with the hardware and software of a commercial CAD system. The accuracy of the segmentation process was determined semiquantitatively and categorized as near perfect if the skin line of the breast was accurately detected, acceptable if only subcutaneous fat was excluded, or unacceptable if any breast parenchyma was excluded from consideration. The accuracy of segmentation was compared for different breast densities and film sizes by using logistic regression (P < .05). RESULTS: Overall, segmentation was near perfect or acceptable in almost 96.8% of images. However, segmentation defects were significantly more common in mammograms with heterogeneously dense breast tissue (8% unacceptable) than in those with fatty replaced (0% unacceptable), scattered (1.2% unacceptable), or extremely dense (1.8% unacceptable) breast parenchyma (P < .05). For images with unacceptable segmentation, the average percentage of breast parenchyma excluded was almost 25% (range, 5%-100%), with no significant differences among breast densities. CONCLUSION: For one commercial CAD system, segmentation was usually near perfect or acceptable but was unacceptable more than five times more frequently for mammograms of breasts with heterogeneously dense parenchyma than for those with all other breast densities. On average, one-quarter of the breast parenchyma was excluded from CAD analysis for images with unacceptable segmentation.

Adult↗

PubMed2005

Computer aid for decision to biopsy breast masses on mammography: validation on new cases.

RATIONALE AND OBJECTIVES: The purpose of this study was to validate the performance of a previously developed computer aid for breast mass classification for mammography on a new, independent database of cases not used for algorithm development. MATERIALS AND METHODS: A computer aid (classifier) based on the likelihood ratio (LRb) was previously developed on a database of 670 mass cases. The 670 cases (245 malignant) from one medical institution were described using 16 features from the American College of Radiology Breast Imaging-Reporting and Data System lexicon and patient history findings. A separate database of 151 (43 malignant) validation cases were collected that were previously unseen by the classifier. These new validation cases were evaluated by the classifier without retraining. Performance evaluation methods included Receiver Operating Characteristic (ROC), round-robin, and leave-one-out bootstrap sampling. RESULTS: The performance of the classifier on the training data yielded an average ROC area of 0.90 +/- 0.02 and partial ROC area (0.90AUC) of 0.60 +/- 0.06. The exact nonparametric performance on the validation set of 151 cases yielded a ROC area of 0.88 and 0.90AUC of 0.57. Using a 100% sensitivity cutoff threshold established on the training data (100% negative predictive value), the classifier correctly identified 100% of the malignant masses in the validation test set, while potentially obviating 26% of the biopsies performed on benign masses. CONCLUSION: The LRb classifier performed consistently on new data that was not used for classifier development. The LRb classifier shows promise as a potential aid in reducing the number of biopsies performed on benign masses.

Biopsy↗

PubMed2005

A framework for optimising the radiographic technique in digital X-ray imaging.

The transition to digital radiology has provided new opportunities for improved image quality, made possible by the superior detective quantum efficiency and post-processing capabilities of new imaging systems, and advanced imaging applications, made possible by rapid digital image acquisition. However, this transition has taken place largely without optimising the radiographic technique used to acquire the images. This paper proposes a framework for optimising the acquisition of digital X-ray images. The proposed approach is based on the signal and noise characteristics of the digital images and the applied exposure. Signal is defined, based on the clinical task involved in an imaging application, as the difference between the detector signal with and without a target present against a representative background. Noise is determined from the noise properties of uniformly acquired images of the background, taking into consideration the absorption properties of the detector. Incident exposure is estimated or otherwise measured free in air, and converted to dose. The main figure of merit (FOM) for optimisation is defined as the signal-difference-to-noise ratio (SdNR) squared per unit exposure or (more preferably) dose. This paper highlights three specific technique optimisation studies that used this approach to optimise the radiographic technique for digital chest and breast applications. In the first study, which was focused on chest radiography with a CsI flat-panel detector, a range of kV(p) (50-150) and filtration (Z = 13-82) were examined in terms of their associated FOM as well as soft tissue to bone contrast, a factor of importance in digital chest radiography. The results indicated that additive Cu filtration can improve image quality. A second study in digital mammography using a selenium direct flat-panel detector indicated improved SdNR per unit exposure with the use of a tungsten target and a rhodium filter than conventional molybdenum target/molybdenum filter techniques. Finally, a third study focusing on cone-beam computed tomography of the breast using a CsI flat-panel detector indicated that high Z filtration of a tungsten target X-ray beam can notably improve the signal and noise characteristics of the image. The general findings highlight the fact that the techniques that are conventionally assumed to be optimum may need to be revisited for digital radiography.

Bone and Bones↗

PubMed2005

Physical characterization of a prototype selenium-based full field digital mammography detector.

The purpose of this study was to measure experimentally the physical performance of a prototype mammographic imager based on a direct detection, flat-panel array design employing an amorphous selenium converter with 70 microm pixels. The system was characterized for two different anode types, a molybdenum target with molybdenum filtration (Mo/Mo) and a tungsten target with rhodium filtration (W/Rh), at two different energies, 28 and 35 kVp, with approximately 2 mm added aluminum filtration. To measure the resolution, the presampled modulation transfer function (MTF) was measured using an edge method. The normalized noise power spectrum (NNPS) was measured by two-dimensional Fourier analysis of uniformly exposed mammograms. The detective quantum efficiencies (DQEs) were computed from the MTFs, the NNPSs, and theoretical ideal signal to noise ratios. The MTF was found to be close to its ideal limit and reached 0.2 at 11.8 mm(-1) and 0.1 at 14.1 mm(-1) for images acquired at an RQA-M2 technique (Mo/Mo anode, 28 kVp, 2 mm Al). Using a tungsten technique (MW2; W/Rh anode, 28 kVp, 2 mm Al), the MTF went to 0.2 at 11.2 mm(-1) and to 0.1 at 13.3 mm(-1). The DQE reached a maximum value of 54% at 1.35 mm(-1) for the RQA-M2 technique at 1.6 microC/kg and achieved a peak value of 64% at 1.75 mm(-1) for the tungsten technique (MW2) at 1.9 microC/kg. Nevertheless, the DQE showed strong exposure and frequency dependencies. The results indicated that the detector offered high MTFs and DQEs, but structured noise effects may require improved calibration before clinical implementation.

Equipment Design↗

PubMed2004

Computer-aided detection in screening mammography: variability in cues.

PURPOSE: To evaluate the variability of true-positive and false-positive cues by using a commercially available computer-aided detection (CAD) system for analysis of 50 malignancies in a screening population. MATERIALS AND METHODS: Fifty breast cancers detected at screening were analyzed by using a commercially available CAD system. Mean patient age was 62.2 years. Each set of mammograms (craniocaudal and mediolateral oblique views) was digitized and analyzed by the CAD system 10 times. One radiologist compared CAD output with the location of the malignancy at mammography and determined whether each lesion was marked accurately in one mammographic view, both views, or neither. Sensitivity and reproducibility of the CAD system were determined for both case- and image-based analysis. RESULTS: Overall sensitivity of the CAD system when at least one of the two mammographic views was marked correctly (case-base sensitivity) was 82.4%. Sensitivity when each mammographic view was considered separately (image-based sensitivity) was 61.1%. For case-based analysis, variability in true-positive CAD cues was demonstrated for 14 of 50 (28%) cases. For image-based analysis, inconsistency in CAD output was observed in 33 of 100 (33%) mammographic views that contained malignancies detected at screening. However, the CAD system consistently detected 40-43 of the 50 breast cancers in each of the 10 CAD runs. Variability for false-positive marks was significantly greater than that for true-positive marks. CONCLUSION: Inconsistency was demonstrated for CAD analysis of breast cancers detected at screening. However, the CAD system was reasonably consistent in the overall number of cancers identified from run to run. Greater variability of the CAD system was also demonstrated for false-positive marks, as compared with true-positive marks.

Adult↗

PubMed2004

Fundamental imaging characteristics of a slot-scan digital chest radiographic system.

Our purpose in this study was to evaluate the fundamental image quality characteristics of a new slot-scan digital chest radiography system (ThoraScan, Delft Imaging Systems/Nucletron, Veenendaal, The Netherlands). The linearity of the system was measured over a wide exposure range at 90, 117, and 140 kVp with added Al filtration. System uniformity and reproducibility were established with an analysis of images from repeated exposures. The modulation transfer function (MTF) was evaluated using an established edge method. The noise power spectrum (NPS) and the detective quantum efficiency (DQE) of the system were evaluated at the three kilo-voltages over a range of exposures. Scatter fraction (SF) measurements were made using a posterior beam stop method and a geometrical chest phantom. The system demonstrated excellent linearity, but some structured nonuniformities. The 0.1 MTF values occurred between 3.3-3.5 mm(-1). The DQE(0.15) and DQE(2.5) were 0.21 and 0.07 at 90 kVp, 0.18 and 0.05 at 117 kVp, and 0.16 and 0.03 at 140 kVp, respectively. The system exhibited remarkably lower SFs compared to conventional full-field systems with anti-scatter grid, measuring 0.13 in the lungs and 0.43 in the mediastinum. The findings indicated that the slot-scan design provides marked scatter reduction leading to high effective DQE (DQEeff) of the system and reduced patient dose required to achieve high image quality.

Equipment Design↗

PubMed2003

Self-organizing map for cluster analysis of a breast cancer database.

The purpose of this study was to identify and characterize clusters in a heterogeneous breast cancer computer-aided diagnosis database. Identification of subgroups within the database could help elucidate clinical trends and facilitate future model building. A self-organizing map (SOM) was used to identify clusters in a large (2258 cases), heterogeneous computer-aided diagnosis database based on mammographic findings (BI-RADS) and patient age. The resulting clusters were then characterized by their prototypes determined using a constraint satisfaction neural network (CSNN). The clusters showed logical separation of clinical subtypes such as architectural distortions, masses, and calcifications. Moreover, the broad categories of masses and calcifications were stratified into several clusters (seven for masses and three for calcifications). The percent of the cases that were malignant was notably different among the clusters (ranging from 6 to 83%). A feed-forward back-propagation artificial neural network (BP-ANN) was used to identify likely benign lesions that may be candidates for follow up rather than biopsy. The performance of the BP-ANN varied considerably across the clusters identified by the SOM. In particular, a cluster (#6) of mass cases (6% malignant) was identified that accounted for 79% of the recommendations for follow up that would have been made by the BP-ANN. A classification rule based on the profile of cluster #6 performed comparably to the BP-ANN, providing approximately 25% specificity at 98% sensitivity. This performance was demonstrated to generalize to a large (2177) set of cases held-out for model validation.

Adult↗

PubMed2003

Application of likelihood ratio to classification of mammographic masses; performance comparison to case-based reasoning.

The likelihood ratio (LR) is an optimal approach for deciding which of two alternate hypotheses best describes a given situation. We adopted this formalism for predicting whether biopsy results of mammographic masses will be benign or malignant, aiming to reduce the number of biopsies performed on benign lesions. We compared the performance of this LR-based algorithm (LRb) to a case-based reasoning (CBR) classifier, which provides a solution to a new problem using past similiar cases. Each classifier used mammographers' BI-RADS descriptions of mammographic masses as input. The database consisted of 646 biopsy-proven mammography cases. Performance was evaluated using Receiver Operating Characteristic (ROC) analysis, Round Robin sampling, and bootstrap. The ROC areas (AUC) for the LRb and CBR were 0.91+/- 0.01 and 0.92 +/- 0.01, respectively. The partial ROC area index (0.90AUC) was the same for both classifiers, 0.59 +/- 0.05. At a sensitivity of 98%, the CBR would spare 204 (49%) of benign lesions from biopsy; the LRb would spare 209 (51%) benign lesions. The performance of the two classifiers was very similar, with no statistical differences in AUC or 0.90AUC. Although the CBR and LRb originate from different fields of study, their implementations differ only in the estimation of the probability density functions (PDFs) of the feature distributions. The CBR performs this estimation implicitly, while using various similarity metrics. On the other hand, the estimation of the PDFs is specified explicitly in the LRb implementation. This difference in the estimation of the PDFs results in the very small difference in performance, and at 98% sensitivity, both classifiers would spare about half of the benign mammographic masses from biopsy. The CBR and LRb are equivalent methods in implementation and performance.

Adult↗

PubMed2003

Computer-aided detection (CAD) in screening mammography: sensitivity of commercial CAD systems for detecting architectural distortion.

OBJECTIVE: Computer-aided detection (CAD) algorithms have successfully revealed breast masses and microcalcifications on screening mammography. The purpose of our study was to evaluate the sensitivity of commercially available CAD systems for revealing architectural distortion, the third most common appearance of breast cancer. MATERIALS AND METHODS: Two commercially available CAD systems were used to evaluate screening mammograms obtained in 43 patients with 45 mammographically detected regions of architectural distortion. For each CAD system, we determined the sensitivity for revealing architectural distortion on at least one image of the two-view mammographic examination (case sensitivity) and for each individual mammogram (image sensitivity). Surgical biopsy results were available for each case of architectural distortion. RESULTS: Architectural distortion was deemed present and actionable by a panel of expert breast imagers in 80 views of the 45 cases. One CAD system detected distortion in 22 of 45 cases of distortion (case sensitivity, 49%) and in 30 of 80 mammograms (image sensitivity, 38%); it displayed 0.7 false-positive marks per image. Another CAD system identified distortion in 15 of 45 cases (case sensitivity, 33%) and 17 of 80 mammograms (image sensitivity, 21%); it displayed 1.27 false-positive marks per image. Sensitivity for malignancy-caused distortion was similar to or lower than sensitivity for all causes of distortion. CONCLUSION: Fewer than one half of the cases of architectural distortion were detected by the two most widely available CAD systems used for interpretations of screening mammograms. Considerable improvement in the sensitivity of CAD systems is needed for detecting this type of lesion. Practicing breast imagers who use CAD systems should remain vigilant for architectural distortion.

Adult↗

PubMed2002

Perceptron error surface analysis: a case study in breast cancer diagnosis.

Perceptrons are typically trained to minimize mean square error (MSE). In computer-aided diagnosis (CAD), model performance is usually evaluated according to other more clinically relevant measures. The purpose of this study was to investigate the relationship between MSE and the area (A(z)) under the receiver operating characteristic (ROC) curve and the high-sensitivity partial ROC area ((0.90)A'(z)). A perceptron was used to predict lesion malignancy based on two mammographic findings and patient age. For each performance measure, the error surface in weight space was visualized. Comparison of the surfaces indicated that minimizing MSE tended to maximize A(z), but not (0.90)A'(z).

Breast↗

PubMed2002

Outcome analysis of patients with acute pancreatitis by using an artificial neural network.

RATIONALE AND OBJECTIVES: The authors performed this study to evaluate the ability of an artificial neural network (ANN) that uses radiologic and laboratory data to predict the outcome in patients with acute pancreatitis. MATERIALS AND METHODS: An ANN was constructed with data from 92 patients with acute pancreatitis who underwent computed tomography (CT). Input nodes included clinical, laboratory, and CT data. The ANN was trained and tested by using a round-robin technique, and the performance of the ANN was compared with that of linear discriminant analysis and Ranson and Balthazar grading systems by using receiver operating characteristic analysis. The length of hospital stay was used as an outcome measure. RESULTS: Hospital stay ranged from 0 to 45 days, with a mean of 8.4 days. The hospital stay was shorter than the mean for 62 patients and longer than the mean for 30. The 23 input features were reduced by using stepwise linear discriminant analysis, and an ANN was developed with the six most statistically significant parameters (blood pressure, extent of inflammation, fluid aspiration, serum creatinine level, serum calcium level, and the presence of concurrent severe illness). With these features, the ANN successfully predicted whether the patient would exceed the mean length of stay (Az = 0.83 +/- 0.05). Although the Az performance of the ANN was statistically significantly better than that of the Ranson (Az = 0.68 +/- 0.06, P < .02) and Balthazar (Az = 0.62 +/- 0.06, P < .003) grades, it was not significantly better than that of linear discriminant analysis (Az = 0.82 +/- 0.05, P = .53). CONCLUSION: An ANN may be useful for predicting outcome in patients with acute pancreatitis.

Acute Disease↗

PubMed2002

Parameter optimization of a computer-aided diagnosis scheme for the segmentation of microcalcification clusters in mammograms.

Our purpose in this study is to develop a parameter optimization technique for the segmentation of suspicious microcalcification clusters in digitized mammograms. In previous work, a computer-aided diagnosis (CAD) scheme was developed that used local histogram analysis of overlapping subimages and a fuzzy rule-based classifier to segment individual microcalcifications, and clustering analysis for reducing the number of false positive clusters. The performance of this previous CAD scheme depended on a large number of parameters such as the intervals used to calculate fuzzy membership values and on the combination of membership values used by each decision rule. These parameters were optimized empirically based on the performance of the algorithm on the training set. In order to overcome the limitations of manual training and rule generation, the segmentation algorithm was modified in order to incorporate automatic parameter optimization. For the segmentation of individual microcalcifications, the new algorithm used a neural network with fuzzy-scaled inputs. The fuzzy-scaled inputs were created by processing the histogram features with a family of membership functions, the parameters of which were automatically extracted from the distribution of the feature values. The neural network was trained to classify feature vectors as either positive or negative. Individual microcalcifications were segmented from positive subimages. After clustering, another neural network was trained to eliminate false positive clusters. A database of 98 images provided training and testing sets to optimize the parameters and evaluate the CAD scheme, respectively. The performance of the algorithm was evaluated with a FROC analysis. At a sensitivity rate of 93.2%, there was an average of 0.8 false positive clusters per image. The results are very comparable with those taken using our previously published rule-based method. However, the new algorithm is more suited to generalize its performance on a larger population, depends on two monotonic outputs making its evaluation much easier and can be trained in an automatic way making practical its application on a large database.

Algorithms↗

PubMed2002

Differences between computer-aided diagnosis of breast masses and that of calcifications.

PURPOSE: To compare the performance of a computer-aided diagnosis (CAD) system for diagnosis of previously detected lesions, based on radiologist-extracted findings on masses and calcifications. MATERIALS AND METHODS: A feed-forward, back-propagation artificial neural network (BP-ANN) was trained in a round-robin (leave-one-out) manner to predict biopsy outcome from mammographic findings (according to the Breast Imaging Reporting and Data System) and patient age. The BP-ANN was trained by using a large (>1,000 cases) heterogeneous data set containing masses and microcalcifications. The performances of the BP-ANN on masses and microcalcifications were compared with use of receiver operating characteristic analysis and a z test for uncorrelated samples. RESULTS: The BP-ANN performed significantly better on masses than microcalcifications in terms of both the area under the receiver operating characteristic curve and the partial receiver operating characteristic area index. A similar difference in performance was observed with a second model (linear discriminant analysis) and also with a second data set from a similar institution. CONCLUSION: Masses and calcifications should be considered separately when evaluating CAD systems for breast cancer diagnosis.

Adult↗

PubMed2002

Cross-institutional evaluation of BI-RADS predictive model for mammographic diagnosis of breast cancer.

OBJECTIVE: Given a predictive model for identifying very likely benign breast lesions on the basis of Breast Imaging Reporting and Data System (BI-RADS) mammographic findings, this study evaluated the model's ability to generalize to a patient data set from a different institution. MATERIALS AND METHODS: The artificial neural network model underwent three trials: it was optimized over 500 biopsy-proven lesions from Duke University Medical Center or "Duke," evaluated on 1,000 similar cases from the University of Pennsylvania Health System or "Penn," and reoptimized for Penn. RESULTS: Trial A's Duke-only model yielded 98% sensitivity, 36% specificity, area index (A(z)) of 0.86, and partial A(z) of 0.51. The cross-institutional trial B yielded 96% sensitivity, 28% specificity, A(z) of 0.79, and partial A(z) of 0.28. The decreases were significant for both A(z) (p = 0.017) and partial A(z) (p < 0.001). In trial C, the model reoptimized for the Penn data yielded 96% sensitivity, 35% specificity, A(z) of 0.83, and partial A(z) of 0.32. There were no significant differences compared with trial B for specificity (p = 0.44) or partial A(z) (p = 0.46), suggesting that the Penn data were inherently more difficult to characterize. CONCLUSION: The BI-RADS lexicon facilitated the cross-institutional test of a breast cancer prediction model. The model generalized reasonably well, but there were significant performance decreases. The cross-institutional performance was encouraging because it was not significantly different from that of a reoptimized model using the second data set at high sensitivities. This study indicates the need for further work to collect more data and to improve the robustness of the model.

Breast Neoplasms↗