Chris Yuzheng Wu

Image feature analysis for classification of microcalcifications in digital mammography: neural networks and genetic algorithms

We have developed an image feature-based algorithm to classify microcalcifications associated with benign and malignant processes in digital mammograms for the diagnosis of breast cancer. The feature-based algorithm is an alternative approach to image based method for classification of microcalcifications in digital mammograms. Microcalcifications can be characterized by a number of quantitative variables describing the underling key features of a suspicious region such as the size, shape, and number of microcalcifications in a cluster. These features are calculated by an automated extraction scheme for each of the selected regions. The features are then used as input to a backpropagation neural network to make a decision regarding the probability of malignancy of a selected region. The initial selection of image features set is a rough estimation that may include redundant and non-discriminant features. A genetic algorithm is employed to select an optimal image feature set from the initial feature set and select an optimized structure of the neural network for the optimal input features. The performance of neural network is compared with that of radiologists in classifying the clusters of microcalcifications. Two set of mammogram cases are used in this study. The first set is from the digital mammography database from the Mammographic Image Analysis Society (MIAS). The second set is from cases collected at Georgetown University Medical Center (GUMC). The diagnostic truth of the cases have been verified by biopsy. The performance of the neural network system is evaluated by ROC analysis. The system of neural network and genetic algorithms improves performance of our previous TRBF neural network. The neural network system was able to classify benign and malignant microcalcifications at a level favorably compared to experienced radiologists. The use of the neural network system can be used to help radiologists reducing the number biopsies in clinical applications. Genetic algorithms are an effective tool to select optimal input features and structure of a backpropagation neural network. The neural network, combined with genetic algorithms, is able to effectively classify benign and malignant microcalcifications. The results of the neural network system can be used to help reducing the number of benign biopsies.

Digital mammography: tradeoffs between 50- and 100-micron pixel size

Improvements in mammography equipment related to a decrease in pixel size of digital mammography detectors raise questions of the possible effects of these new detectors. Mathematical modeling suggested that the benefits of moving from 100 to 50 micron detectors were slight and might not justify the cost of these new units. Experiments comparing screen film mammography, a storage phosphor 100 micron digital detector, a 50 micron digital breast spot device, 100 micron film digitization and 50 micron film digitization suggests that object conspicuity should be better for digital compared to conventional systems, but that there seemed to be minimal advantage to going from 100 to 50 microns. The 50 micron pixel system appears to provide a slight advantage in object contrast and perhaps in shape definition, but did not allow smaller objects to be detected.

Classification of microcalcifications in radiographs of pathological specimen for the diagnosis of breast cancer

A convolution neural network (CNN) was employed to classify benign and malignant microcalcifications in the radiographs of pathological specimen. The input signals to the CNN were the pixel values of image blocks centered on each of the suspected microcalcifications. The CNN has been shown to be capable of recognizing different image patterns. Digital images were acquired by digitizing radiographs at a high resolution of 21 micrometers X 21 micrometers . Eighty regions of interest (ROIs) selected from digitized radiographs of pathological specimen were used for the training and testing of the neural network system. The performance of the neural network system was analyzed using the ROC analysis.

Diagnosis of breast cancer by MRI: a 3D computer visualization and analysis system

The conventional diagnosis of breast cancer by a combination of mammography and physical examination has limited efficacy. Only 20 - 30% of the suspicious breast lesions biopsied are actually malignant. Breast MRI (BMRI), using intravenous contrast injection to detect and characterize breast lesions, has shown promise in several studies. We are developing a 3D image visualization and analysis system to assist radiologists to detect breast cancer in BMRI. Dynamic contrast-enhanced MRI has emerged to become an effective procedure in BMRI for the diagnosis and management of breast carcinoma. A 3D image visualization and analysis system that allows a radiologist to rapidly search through BMRI images and visualize three- dimensional volume of a whole breast has been developed. Three dimensional breast images were constructed from 2D slices. The system can register and subtract a pre-contrast image from each of the time sequenced post-contrast images. The dynamic time sequences of the breast before and after the contrast administration can be visualized. Suspicious lesions can be detected based on the dynamic time-sequence changes in images. The system also allows interactive manipulation of images for viewing from different angles and examination of 2D projections at specific locations. The developed 3D image visualization and analysis system provided radiologists an efficient way to analyze MR breast images. Our automated detection scheme has the potential to accurately detect suspicious lesions and can be an effective tool to facilitate the clinical application of MRI for breast imaging.

Digital mammography: the effects of decreased exposure

It has been stated that digital mammography will reduce the exposure required for mammography. This poster explores the effects of decreased exposure on the information present in digital mammography. In general, the digital system performed better than screen film mammography with lower exposures. With the usual exposures used for screen film mammography, performance was equal. With high exposures sufficient to result in a dark film (OD 1.5), the digital system performed better than screen film with very small test objects. Proposals have been made to decrease the tube loading required for slot scanning devices by increasing KVP. This would result in their being less object contrast due to the decreases in the absorption coefficient of calcium compared to water at higher KVP. This poster looks at the potential for correcting the loss in object contrast that would result from the use of high contrast look up tables. It was found that in the tested system, one could restore the information in one of the two test objects used (but not the other), but that the image processing methods used would result in an image that radiologists would probably find inadequate for interpretation.

Adaptive-sized neural-networks-based computer-aided diagnosis of microcalcifications

In this report, we present an adaptive-sized neural network model for the detection of microcalcifications. The neural network has capabilities of automatically adjusting the network size depending on the training set, of rejecting unknown inputs, and of fast learning. When the adaptive-sized neural network is used, the user can find the optimal network size without trial and error. In addition, the reliability of the network performance is high because of the rejection of unlearned inputs. The inputs for the neural network used in this study were 11 X 11 pixel sub-images that were extracted from digitized mammograms. The experiments in 83.3% sensitivity, 84.3% specificity, and 22.4% rejection rate. The weight patterns after learning process and the dependency of the network performance on the order of presenting training examples were also studied.

Quality control in digital mammography: automatic detection of under- and over-exposed mammograms

We developed a quality control system (QCS) for digital mammography that can notify technologists in real time of mammograms of poor image quality due to under or over exposure. Mammograms are digitized by a Lumisys Scanner at 100 micron and 12 bits per pixel. An automatic image segmentation technique is employed to extract area inside the breast in mammogram. Histograms of the segmented areas are then calculated. By analyzing the composition of histograms, the computer program determines whether the original films have properly exposed. Traditional image segmentation techniques are based on histogram analysis of digitized mammograms. However, such methods often fail with mammograms of low contrast or that are under-exposed because the difference in brightness across the breast skin line is so small that it is difficult to define boundary by thresholding or region growing techniques. We proposed a novel method to detect breast skin line based on statistical changes of gradient. By analyzing the histogram composition of normal, under and over-exposed films, we defined an image feature that describes the image intensity content of underlying mammograms. The criterion for determining the category of a mammogram were established by studying a training database of normal, under, and over exposed films. We can then classify the mammograms using the image feature, based on the established criterion. Over 150 real mammograms of different exposure levels were analyzed. The images were classified by the computer system into groups of normal, slightly under-exposed, under-exposed, slightly over- exposed, and over-exposed. We compared the classification results by computer with a radiologists evaluation. Our QCS system was able to correctly classify over 85% of the cases. Receiver operating curve (ROC) analysis will be employed to evaluate the performance of the QCS system in determining the image quality of digital mammograms. Our QCS program is able to automatically determine whether a mammogram is properly exposed and advise a technologist to re-take additional exposures. The QCS correctly identified 100% of over- and under-exposed mammograms and 92% of mammograms of normal exposure. The QCS can help reduce the cost of recalling patients and improve the overall quality of mammographic service.

Design-based approach to ethics in computer-aided diagnosis

A design-based approach to ethical analysis examines how computer scientists, physicians and patients make and justify choices in designing, using and reacting to computer-aided diagnosis (CADx) systems. The basic hypothesis of this research is that values are embedded in CADx systems during all phases of their development, not just retrospectively imposed on them. This paper concentrates on the work of computer scientists and physicians as they attempt to resolve central technical questions in designing clinically functional CADx systems for lung cancer and breast cancer diagnosis. The work of Lo, Chan, Freedman, Lin, Wu and their colleagues provides the initial data on which this study is based. As these researchers seek to increase the rate of true positive classifications of detected abnormalities in chest radiographs and mammograms, they explore dimensions of the fundamental ethical principal of beneficence. The training of CADx systems demonstrates the key ethical dilemmas inherent in their current design.

Artificial convolution neural network with wavelet kernels for disease pattern recognition

A two-dimensional convolution neural network (CNN) with wavelet kernels (WK) has been developed for image pattern recognition. The structure of the CNN is a simplified version of the neocognitron. We used only a two-level structure and eliminated all complex-cell layers. Nets between two adjacent layers in the feature selection level of the CNN are selectively interconnected across groups. In this part of the CNN signals processing, each group in the receiving layer receives signals from a group of weights (i.e., kernels). For the forward signal propagation, the product obtained from the kernel convoluting the front layer is collected onto the corresponding matrix element of the receiving layer. In this paper, the convolution kernels of the CNN (CNN/WK) are wavelet based and are trained by a supervised manner. In the development of the CNN/WK, we forced each updated convolution kernel to be orthonormal. Therefore, features (transformed coefficients) selected on the transform domain are linearly independent. Hence, the fully connected layers in the classification level of the CNN can perform more effectively. The applications of the CNN for disease pattern recognition have been very successful. When isolated patterns were further processed by internal filtering and classification layers were built into the neural network structure, the disease patterns were more easily recognized. Although, we did not receive substantial improvement of the ROC performance using the CNN/WK, this method may assist us in the analysis of the trained kernels and eventually lead to the optimization of feature extraction in a course of disease pattern recognition.

Academic consortium for the evaluation of computer-aided diagnosis (CADx) in mammography

Computer aided diagnosis (CADx) is a promising technology for the detection of breast cancer in screening mammography. A number of different approaches have been developed for CADx research that have achieved significant levels of performance. Research teams now recognize the need for a careful and detailed evaluation study of approaches to accelerate the development of CADx, to make CADx more clinically relevant and to optimize the CADx algorithms based on unbiased evaluations. The results of such a comparative study may provide each of the participating teams with new insights into the optimization of their individual CADx algorithms. This consortium of experienced CADx researchers is working as a group to compare results of the algorithms and to optimize the performance of CADx algorithms by learning from each other. Each institution will be contributing an equal number of cases that will be collected under a standard protocol for case selection, truth determination, and data acquisition to establish a common and unbiased database for the evaluation study. An evaluation procedure for the comparison studies are being developed to analyze the results of individual algorithms for each of the test cases in the common database. Optimization of individual CADx algorithms can be made based on the comparison studies. The consortium effort is expected to accelerate the eventual clinical implementation of CADx algorithms at participating institutions.