We have developed a computerized scheme for detection of interstitial lung disease by using artificial neural networks (ANNs) on quantitative analysis of digital image data. Three separate ANNs were applied for the ANN scheme. The first ANN was trained with horizontal profiles in the ROIs selected from digital chest radiographs. The second ANN was trained with vertical output patterns obtained from the Ist ANN in each ROI. The output from the 2nd ANN was used to distinguish between normal and abnormal ROIs. In order to improve the performance, we attempted a density correction and rib edge removal. The A(Z) value was improved from 0.906 to 0.934 by incorporating density correction For the classification of each chest image, we employed a rule-based method and a rule-based plus the third ANN method. A high A(Z) value (0.976) was obtained with the rule-based plus ANN method. The ANNs can learn certain statistical properties associated with patterns of interstitial infiltrates in chest radiographs.

A novel technique for optimizing the wavelet transform to enhance and detect microcalcifications in mammograms was developed based on the supervised learning method. In the learning process, a cost function is formulated to represent the difference between a desired output and the reconstructed image obtained from weighted wavelet coefficients for a given mammogram. This cost function is then minimized by modifying the weights for wavelet coefficients via a conjugate gradient algorithm. The Least Asymmetric Daubechies' wavelets were optimized with 44 regions-of-interest as the training set using a jackknife method. The performance of the optimized wavelets achieved a sensitivity of 90% with specificity of 80%, which outperforms the authors' current scheme based on a conventional wavelet transform.

Artificial neural networks have been applied to the differentiation of actual "true" clusters from normal parenchymal patterns and also to the differentiation of actual clusters from false-positive clusters as reported by a computerized scheme for the detection of microcalcifications in digital mammograms. The differentiation was carried out in both the spatial and frequency domains. The performance of the neural networks was evaluated quantitatively by means of receiver operating characteristic (ROC) analysis. It was found that the networks could distinguish clustered microcalcifications from normal nonclustered areas in the frequency domain, and that they could eliminate approximately 50% of false-positive clusters of microcalcifications while preserving 95% of the positive clusters, when applied to the results of the automated detection scheme. A large, comprehensive training database is needed for neural networks to perform reliably in clinical situations.

At present mammography is the most effective method for the early detection of breast cancer1 . Detection and classification of masses in mammograms are among the most important and difficult tasks performed by radiologists. Various studies have indicated that regular mammographic screening can reduce the mortality from breast cancer in women2. Thus mammography may become one of the largest volume x-ray procedures routinely interpreted by radiologists. The miss rate for the radiographic detection of malignant masses ranges from 12 to 30 percent. In addition although general rules exist for the visual differentiation of benign and malignant masses error does occur in the classification of masses with the current methods of radiologic characterization. Thus it is apparent that the efficiency and effectiveness of screening procedures could be increased by use of a computer system that successfully aids the radiologist in detecting and characterizing mammographic masses. We are developing computerized schemes for the automated detection and classification of masses in digital mammograms. The detection scheme utilizes the architectural symmeiry of the left and right breasts and digital bilateralsubtraction techniques in order to increase the conspicuity of the mammographic mass prior to the application of featureextraction techniques. The classification scheme involves the extraction of border information from the mammographic mass in order to quantify the degree of spiculation which is related to the likelihood of malignancy. METHODS Clinical screen/film mammograms were used in the

The authors investigated the feasibility of using computer methods for automated detection of clustered microcalcifications on clinical mammograms. A new difference-image approach using a matched filter/box-rim filter combination effectively removed the structured background from the image. A locally adaptive gray-level thresholding technique was then used for extraction of the signals from the resulting difference image. Signal-extraction criteria based on the size, contrast, number, and clustering properties of microcalcifications were next imposed on the detected signals to distinguish true signals from noise or artifacts. The detection accuracy of the computer scheme was evaluated by means of a free response receiver operating characteristic (FROC) analysis. It was found that, for simulated subtle microcalcifications superimposed on normal mammograms, the difference-image approach with a matched filter/box-rim filter combination could yield a true-positive cluster detection rate of 80% at a false-positive detection rate of one cluster per image. In a study of 20 clinical images containing moderately subtle microcalcifications, the automated computer scheme obtained an 82% true-positive cluster detection rate at a false-positive detection rate of one cluster per image. These results indicate that the automated method has the potential to aid radiologists in screening mammograms for clustered microcalcifications.