Biomedical subjects

Sung-Nien Yu

Publications and source records attributed to Sung-Nien Yu.

4 recordsLinked to original sources

PubMed2006

Detection of microcalcifications in digital mammograms using wavelet filter and Markov random field model.

Clustered microcalcifcations (MCs) in digitized mammograms has been widely recognized as an early sign of breast cancer in women. This work is devoted to developing a computer-aided diagnosis (CAD) system for the detection of MCs in digital mammograms. Such a task actually involves two key issues: detection of suspicious MCs and recognition of true MCs. Accordingly, our approach is divided into two stages. At first, all suspicious MCs are preserved by thresholding a filtered mammogram via a wavelet filter according to the MPV (mean pixel value) of that image. Subsequently, Markov random field parameters based on the Derin-Elliott model are extracted from the neighborhood of every suspicious MCs as the primary texture features. The primary features combined with three auxiliary texture quantities serve as inputs to classifiers for the recognition of true MCs so as to decrease the false positive rate. Both Bayes classifier and back-propagation neural network were used for computer experiments. The data used to test this method were 20 mammograms containing 25 areas of clustered MCs marked by radiologists. Our method can readily remove 1341 false positives out of 1356, namely, 98.9% false positives were removed. Additionally, the sensitivity (true positives rate) is 92%, with only 0.75 false positives per image. From our experiments, we conclude that, with a proper choice of classifier, the texture feature based on Markov random field parameters combined with properly designed auxiliary features extracted from the texture context of the MCs can work outstandingly in the recognition of MCs in digital mammograms.

Algorithms↗

PubMed2005

A three-object model for the similarity searches of chest CT images.

We propose in this paper a three-object model specifically for the archiving and retrieval of chest CT images. To calculate parameters for the model, each chest CT image needs to be processed to segment the three main objects and then the features be extracted to describe the objects' properties and relationships. In the image segmentation part, we applied the knowledge of the modality on chest CT images and modified the traditional watershed image segmentation algorithm including a four-step merging algorithm specifically for chest CT images. After segmentation, the mediastinum and two lung lobes are identified. The mediastinum object is mainly described by shape-related features while the two lung lobes are described mainly by texture features. A three-object model was exploited to describe the object features and the spatial relationship among objects. To test the capability of the three-object model to the similarity searches of chest CT images, we developed a CBIR system in which three distinct query modes were provided. They are 'searching by ARGs', 'searching by shape features of mediastinum', and 'searching by texture features of lung lobes'. The experimental results show that the three-object model demonstrates impressive power in the similarity searching of chest CT images. Among the three searching modes, the 'searching by shape features of mediastinum' and 'searching by texture features of lung lobes' modes provide user choices to search for images with high similarities in specific objects rather than in the whole images. The precision rate of either query mode is high, with an average of around 80% out of the first 30 result images are justified as similar, which is impressive in a fully automatic image query system using content features. Nevertheless, the two query modes that concentrate on distinct object features show slightly better capability in searching for similar images than the 'searching by ARGs' mode.

Humans↗

PubMed2005

Quantitatively characterizing the textural features of sonographic images for breast cancer with histopathologic correlation.

OBJECTIVE: In this study, quantitative characterization of sonographic image texture and its correlation with histopathologic findings was developed for facilitating clinical diagnosis. A statistical feature matrix was applied to quantify the texture difference (ie, the dissimilarity) of the sonographic images for malignant and benign breast tumors. METHODS: Thirty-three patients were recruited for this study. Imaging was performed on a commercially available sonographic imaging system in clinical use. The parameters used for image acquisition were kept the same during clinical examination. RESULTS: On the basis of dissimilarity values, 3 phenomena were noted in the relatively large malignancies studied. First, stellate carcinoma showed the least dissimilarity on sonographic images; second, circumscribed carcinoma showed the most dissimilarity; and third, malignant tissue mixed with fibrous and cellular parts (dense lymphocyte infiltration and prominent intraductal tumors) had dissimilarity values in between. Image textures with smaller dissimilarity values (especially for those values <4.4 in our study) are likely to be stellate carcinoma. CONCLUSIONS: From the experimental results, it is shown that the cellular and fibrous content with spatial distribution of breast masses determine the dissimilarity values on sonographic images. The dissimilarity may be used to quantitatively represent the image texture and is well correlated with the histopathologic description.

Adolescent↗

PubMed2004

An improved Maxnet.

In the proposed model, dynamic inhibitory weights are used to speed up the convergence rate, and a new convergence rule is applied to find all maxima. The hardware implementation of the proposed model is presented in the study, and simulation results indicate that the proposed model converges much faster than the other networks.

Algorithms↗