Improving Performance of Breast Cancer Risk Prediction by Incorporating Optical Density Image Feature Analysis: An Assessment

RATIONALE AND OBJECTIVES

The purpose of this study is to improve accuracy of near-term breast cancer risk prediction by applying a new mammographic image conversion method combined with a two-stage artificial neural network (ANN)-based classification scheme.

MATERIALS AND METHODS

The dataset included 168 negative mammography screening cases. In developing and testing our new risk model, we first converted the original grayscale value (GV)-based mammographic images into optical density (OD)-based images. For each case, our computer-aided scheme then computed two types of image features representing bilateral asymmetry and the maximum of the image features computed from GV and OD images, respectively. A two-stage classification scheme consisting of three ANNs was developed. The first stage included two ANNs trained using features computed separately from GV and OD images of 138 cases. The second stage included another ANN to fuse the prediction scores produced by two ANNs in the first stage. The risk prediction performance was tested using the rest 30 cases.

RESULTS

With the two-stage classification scheme, the computed area under the receiver operating characteristic curve (AUC) was  0.816 ± 0.071, which was significantly higher than the AUC values of 0.669 ± 0.099 and 0.646 ± 0.099 achieved using two ANNs trained using GV features and OD features, respectively (P < .05).

CONCLUSION

This study demonstrated that applying an OD image conversion method can acquire new complimentary information to those acquired from the original images. As a result, fusion image features computed from these two types of images yielded significantly higher performance in near-term breast cancer risk prediction.

Development of breast lesions models database

PURPOSE

We present the development and the current state of the MaXIMA Breast Lesions Models Database, which is intended to provide researchers with both segmented and mathematical computer-based breast lesion models with realistic shape.

METHODS

The database contains various 3D images of breast lesions of irregular shapes, collected from routine patient examinations or dedicated scientific experiments. It also contains images of simulated tumour models. In order to extract the 3D shapes of the breast cancers from patient images, an in-house segmentation algorithm was developed for the analysis of 50 tomosynthesis sets from patients diagnosed with malignant and benign lesions. In addition, computed tomography (CT) scans of three breast mastectomy cases were added, as well as five whole-body CT scans. The segmentation algorithm includes a series of image processing operations and region-growing techniques with minimal interaction from the user, with the purpose of finding and segmenting the areas of the lesion. Mathematically modelled computational breast lesions, also stored in the database, are based on the 3D random walk approach.

RESULTS

The MaXIMA Imaging Database currently contains 50 breast cancer models obtained by segmentation of 3D patient breast tomosynthesis images; 8 models obtained by segmentation of whole body and breast cadavers CT images; and 80 models based on a mathematical algorithm. Each record in the database is supported with relevant information. Two applications of the database are highlighted: inserting the lesions into computationally generated breast phantoms and an approach in generating mammography images with variously shaped breast lesion models from the database for evaluation purposes. Both cases demonstrate the implementation of multiple scenarios and of an unlimited number of cases, which can be used for further software modelling and investigation of breast imaging techniques. The created database interface is web-based, user friendly and is intended to be made freely accessible through internet after the completion of the MaXIMA project.

CONCLUSIONS

The developed database will serve as an imaging data source for researchers, working on breast diagnostic imaging and on improving early breast cancer detection techniques, using existing or newly developed imaging modalities.

A new near-term breast cancer risk prediction scheme based on the quantitative analysis of ipsilateral view mammograms

PURPOSE

To help improve efficacy of screening mammography and eventually establish an optimal personalized screening paradigm, this study aimed to develop and test a new near-term breast cancer risk prediction scheme based on the quantitative analysis of ipsilateral view of the negative screening mammograms.

METHODS

The dataset includes digital mammograms acquired from 392 women with two sequential full-field digital mammography examinations. All the first ("prior") sets of mammograms were interpreted as negative during the original reading. In the sequential ("current") screening, 202 were proved positive and 190 remained negative/benign. For each pair of the "prior" ipsilateral mammograms, we adaptively fused the image features computed from two views. Using four different types of image features, we built four elastic net support vector machine (EnSVM) based classifiers. Then, the initial prediction scores form the 4 EnSVMs were combined to build a final artificial neural network (ANN) classifier that produces the final risk prediction score. The performance of the new scheme was evaluated by using a 10-fold cross-validation method and an assessment index of the area under the receiver operating characteristic curve (AUC).

RESULTS

A total number of 466 features were initially extracted from each pair of ipsilateral mammograms. Among them, 51 were selected to build the EnSVM based prediction scheme. The AUC = 0.737 ± 0.052 was yielded using the new scheme. Applying an optimal operating threshold, the prediction sensitivity was 60.4% (122 of 202) and the specificity was 79.0% (150 of 190).

CONCLUSION

The study results showed moderately high positive association between computed risk scores using the "prior" negative mammograms and the actual outcome of the image-detectable breast cancers in the next subsequent screening examinations. The study also demonstrated that quantitative analysis of the ipsilateral views of the mammograms enabled to provide useful information in predicting near-term breast cancer risk.

Applying a new bilateral mammographic density segmentation method to improve accuracy of breast cancer risk prediction

PURPOSE

How to optimally detect bilateral mammographic asymmetry and improve risk prediction accuracy remains a difficult and unsolved issue. Our aim was to find an effective mammographic density segmentation method to improve accuracy of breast cancer risk prediction.

METHODS

A dataset including 168 negative mammography screening cases was used. We applied a mutual threshold to bilateral mammograms of left and right breasts to segment the dense breast regions. The mutual threshold was determined by the median grayscale value of all pixels in both left and right breast regions. For each case, we then computed three types of image features representing asymmetry, mean and the maximum of the image features, respectively. A two-stage classification scheme was developed to fuse the three types of features. The risk prediction performance was tested using a leave-one-case-out cross-validation method.

RESULTS

By using the new density segmentation method, the computed area under the receiver operating characteristic curve was 0.830 ± 0.033 and overall prediction accuracy was 81.0%, significantly higher than those of 0.633 ± 0.043 and 57.1% achieved by using the previous density segmentation method ([Formula: see text], t-test).

CONCLUSIONS

A new mammographic density segmentation method based on a bilateral mutual threshold can be used to more effectively detect bilateral mammographic density asymmetry and help significantly improve accuracy of near-term breast cancer risk prediction.

Automatic feature learning using multichannel ROI based on deep structured algorithms for computerized lung cancer diagnosis

This study aimed to analyze the ability of extracting automatically generated features using deep structured algorithms in lung nodule CT image diagnosis, and compare its performance with traditional computer aided diagnosis (CADx) systems using hand-crafted features. All of the 1018 cases were acquired from Lung Image Database Consortium (LIDC) public lung cancer database. The nodules were segmented according to four radiologists' markings, and 13,668 samples were generated by rotating every slice of nodule images. Three multichannel ROI based deep structured algorithms were designed and implemented in this study: convolutional neural network (CNN), deep belief network (DBN), and stacked denoising autoencoder (SDAE). For the comparison purpose, we also implemented a CADx system using hand-crafted features including density features, texture features and morphological features. The performance of every scheme was evaluated by using a 10-fold cross-validation method and an assessment index of the area under the receiver operating characteristic curve (AUC). The observed highest area under the curve (AUC) was 0.899±0.018 achieved by CNN, which was significantly higher than traditional CADx with the AUC=0.848±0.026. The results from DBN was also slightly higher than CADx, while SDAE was slightly lower. By visualizing the automatic generated features, we found some meaningful detectors like curvy stroke detectors from deep structured schemes. The study results showed the deep structured algorithms with automatically generated features can achieve desirable performance in lung nodule diagnosis. With well-tuned parameters and large enough dataset, the deep learning algorithms can have better performance than current popular CADx. We believe the deep learning algorithms with similar data preprocessing procedure can be used in other medical image analysis areas as well.

Computer-assisted detection of swallowing difficulty

To evaluate classification performance of a support vector machine (SVM) classifier for diagnosing swallowing difficulty based on the hyoid movement data attained from videofluoroscopic swallowing study, the hyoid kinematics during the swallowing of 2 mL of liquid barium solution were analyzed for 90 healthy volunteers and 116 dysphagic stroke patients. SVM was used to classify the kinematic results as normal or dysfunctional swallowing. Various kernel functions and kernel parameters were used for optimization. Features were selected to find an optimal feature subset and to minimize redundancy. Accuracy, sensitivity, specificity, and area under a receiving operating characteristic curve (AUC) were used to assess the discrimination performance. In 19 out of 26 features, mean comparison revealed a significant difference between healthy subjects and dysphagic patients. By reducing the number of features to 10, an AUC of 0.9269 could be reached. Common features showing the best classification in both kernel functions included forward maximum excursion time, upward maximum excursion time, maximum excursion length, upward maximum velocity time, upward maximum acceleration time, maximum acceleration, maximum acceleration time, and mean acceleration. SVM-based classification method with the use of kernel functions showed an outstanding (AUC of 0.9269) discrimination performance for either healthy or dysphagic hyoid movement during swallowing. We expect that this classification method will be useful as an adjunct diagnostic tool by providing automatic detection of swallowing dysfunction as well as a research tool providing deeper understanding of pathophysiology.