Brian Krasner

A Comparison of Case Retrieval Times: Film Versus Picture Archiving and Communications Systems

One of the advantages that a picture archiving and communications system (PACS) is supposed to provide over a film-based operation is improved performance in retrieving images. Although it seems self-evident that this should be so, this experiment was intended to verify this and to provide some time comparisons for the two methods. The experiment consisted of randomly selecting ultrasound and computed tomography cases and determining how long it took to retrieve files at a PACS workstation or in person from the file room. To simulate actual retrieval volumes, a total of 40 cases from current to 6 months old, 20 cases from the past year, and 10 cases more than 1 year old was selected. Results indicate that PACS retrieval can indeed be faster than file room retrieval. However, the difference is less for recent cases than for older cases. For cases 6 or fewer months old, the workstation retrieval was approximately 2.5 minutes faster per case than the film file room. This time difference increased markedly when extended to the 1-year and older-than-1-year groups. This report details the results of this study and provides information about the reliability of the two archives.

Full-frame entropy coding for radiological image compression

The discrete cosine transform (DCT) type algorithm is a leading technique in irreversible image compression. The full-frame bit-allocation (FFBA) technique based on DCT has been proven to be an outstanding compression method for radiological image. A new version of this technique, which discards the bit-table and uses entropy coding instead of bit-packing, FFEC, has recently been developed. The results showed that the new method improves compression ratio by a factor of two (e.g., 24:1 vs. 13:1) with a controlled mean-square-error. With the same compression ratio, the FFEC produces about 50 less error than does FFBA. The FFEC has been tested and results are demonstrated in this paper along with FFBA and the splitting and remapping method. This new method is characterized as a simple, error controllable, and highly efficient compression technique.

Effect of vector quantization on ultrasound tissue characterization

This paper presents a case study of the effects of compression error on computerized tissue characterization of normal and fatty ultrasound liver images. Two compression techniques were studied, pruned-tree structured vectored quantization (PTSVQ) and PTSVQ with splitting. Vector quantization is a technique for representing a block of image values, or vector, by the vector in a codebook that is closest to the original vector. Splitting is a technique for decomposing image pixel values into the high and low values. The high values are compressed reversibly while the low values are compressed via PTSVQ. Tissue characterization was accomplished by extracting features from a region of interest (ROI). These features included measuring fractal dimension and statistics concerning run length and co-occurrence probabilities of pixels separated by a given direction and distance. The results were: (1) PTSVQ with splitting produced less image distortion at moderate bit rates than PTSVQ as measured by mean square error; (2) PTSVQ with splitting produced more degradation of the tissue characterizer; and (3) Rotation of the ROIs greatly reduced the degradation of the tissue characterizer for both types of compression. This type of rotation uses interpolation to derive pixel values for rotated lattice points that fall between original lattice points. A possible explanation for these results is that PTSVQ caused irregular distortions at edges depending upon the amount of region information included in the design of the codebook. The interpolation during rotation reduces these irregularities.

Optimization of quantitative sonographic diagnostic analysis of breast lesions

The purpose of this study was to enhance the ability of quantitative sonography to distinguish between B-scan images of malignant and benign lesions of the breast. Several second-order pixel gray level statistics have been used to achieve a good but not acceptable diagnostic accuracy in characterizing breast lesions. Therefore, this study sought to optimize the diagnostic accuracy of second order statistics. The co-occurrence matrix is the most useful second-order statistic so far studied. It is an estimate of the joint probability distribution of gray levels of two pixels separated by a given distance and orientation. Several distances and orientations have been tried previously, but no systematic attempt had been made to find the optimum parameters for diagnosis. In this study, co-occurrence statistics of malignant and benign lesion images were determined as a function of distance and orientation. In particular, the correlation function was modeled as a separable, exponential function, first order for increments in both the x and y directions. Model parameters were used as features for discriminating benign from cancer lesions. An attempt was made to optimize the features by excluding the noisy data from the fit and again using the model parameters.

Computer-assisted diagnosis for lung nodule detection using a neural network technique

The potential advantages of using digital techniques instead of film-based radiology have been discussed very extensively for the past ten years. These advantages are found mainly in the computer management of picture archiving and communication systems (PACS). On the other hand, the computer-assisted diagnosis (CADx) could potentially enhance radiological services in the future. Lung nodule detection has been a clinically difficult subject for many years. Most of the literature has indicated that the finding rate for lung nodules (size range from 3 mm to 15 mm) is only about 65%, and 30% of the missing nodule can be found retrospectively. In the recent research, imaging processing techniques, such as thresholding and morphological analysis, have been employed to enhance the true-positive detection. However, these methods still produce many false-positive detections. We have used neural networks to distinguish true-positives from the suspected areas-of-interest which are generated from signal enhanced image. The initial results show that the trained neural networks program can increase true-positive detections and drastically reduce the number of false-positive detections. This program can perform three modes of lung nodule detection: (1) thresholding, (2) profile matching analysis, and (3) neural network. This program is fully automatic and has been implemented in a DEC 5000/200 workstation. The total processing time for all three methods is less than 35 seconds. We are planning to link this workstation to our PACS for further clinical evaluation. In this paper, we report our neural network and fast algorithms for various image processing techniques for the lung nodule detection and show the results of the initial studies.

Hybrid coding for split gray values in radiological image compression

Digital techniques are used more often than ever in a variety of fields. Medical information management is one of the largest digital technology applications. It is desirable to have both a large data storage resource and extremely fast data transmission channels for communication. On the other hand, it is also essential to compress these data into an efficient form for storage and transmission. A variety of data compression techniques have been developed to tackle a diversity of situations. A digital value decomposition method using a splitting and remapping method has recently been proposed for image data compression. This method attempts to employ an error-free compression for one part of the digital value containing highly significant value and uses another method for the second part of the digital value. We have reported that the effect of this method is substantial for the vector quantization and other spatial encoding techniques. In conjunction with DCT type coding, however, the splitting method only showed a limited improvement when compared to the nonsplitting method. With the latter approach, we used a nonoptimized method for the images possessing only the top three-most-significant- bit value (3MSBV) and produced a compression ratio of approximately 10:1. Since the 3MSB images are highly correlated and the same values tend to aggregate together, the use of area or contour coding was investigated. In our experiment, we obtained an average error-free compression ratio of 30:1 and 12:1 for 3MSB and 4MSB images, respectively, with the alternate value contour coding. With this technique, we clearly verified that the splitting method is superior to the nonsplitting method for finely digitized radiographs.

Radiological image compression: image characteristics and clinical considerations

Since most medical images are composed of different image characteristics, it has been demonstrated that applying a combined method to various image components can preserve a higher image quality. The major image components are: (a) smooth areas, (b) sharp edges, (c) texture, and (d) noise. In practice, sharp edges and general textures are two main components to be concerned for radiological images compression. A unified perspective of transform coding is reviewed to find out how a high compression ratio can be achieved with a lossy compression technique. Theoretically, the image quality in resolution power is associated with a composed module transfer function (MTF) when an image obtained from x-ray device coupling with a digitization module and processed by a lossy compression. It is very difficult to use a global MTF (or a band of MTF) to represent such a system. In this paper, we concentrate on clinical considerations for various applications in radiological image compression. Three different applications and associated compression strategies are discussed. Based on these compression strategies, we believe that many compression methods are suitable for clinical implementation with some clinical guidance and technical modifications.

Narrow bandwidth spectral analysis of the textures of interstitial lung diseases

The object of this study was to develop a classifier for distinguishing between regions-of-interest (ROIs) from normal lung radiographs and ROIs from radiographs showing interstitial lung disease. The method used was to estimate the covariance statistics of the ROIs of the lung interstitial space and, based on the estimate, to design filters for isolating statistically significant components of the spectrum. The energy of filtered images was used as a classifier. Additionally, the filtered images were analyzed and classified using a convolution neural network (CNN). The procedure used to generate the filters was: (1) Convert 2D neighborhoods of pixels to vectors. (2) Form the sample covariance matrix from the vectors. (3) Compute the eigenvectors and eigenvalues of the matrix. (4) Convert the eigenvectors back to 2D form and use as filters. The images selected for study included normal lungs, and lungs with different types and profusions of pneumoconiosis opacities. One group of ROIs of the interstitial space was used to design filters. Another group was used as a test of classification accuracy. The results showed that the designed classifier was effective in discriminating ROIs with small pneumoconiosis opacities from normal ROIs.

Gain of using irreversible over error-free data compression in digital radiography

In our previous studies we found that excessive spatial resolution can be partially recovered by error-free compression. However, the excessive gray dynamic resolution is not reducible as far as information content is concerned. Both information evaluation and preliminary clinical tests indicated that no information existed beyond 9-bit when digitized by a sample size of 180 (mu) with laser film digitization or computed radiography. In this study, we found that a high compression method can only achieve a small fraction of true compression efficacy over an error-free compression for a well-defined digital radiographic imaging system. This implies that the following two procedures contain similar digital information: (a) digitization of a 12- bit image and processing by a moderate irreversible compression (e.g., DCT type compression) and (b) digitization of the same image 8-bit followed by an error-free compression method (e.g., DPCM/arithmetic coding). Images processed by the above methods require about the same digital storage. The image quality of 12-bit with 5:1 irreversible compression is very close to that of 8-bit with 3:1 error-free compression. Higher compression efficiency (e.g., 0.5 bit/pixel) using procedure (a) would degrade the image quality particularly in edges and small structures. This is because the quantization procedure acts as a filter in the DCT compression. Without the interference of noise, the compression efficiency of using irreversible and reversible compression techniques are comparable as far as information is concerned.

Vector quantization distortion of medical ultrasound features.

Pruned-tree structured vectored quantization (PTSVQ) was applied to the lower five gray scale remapped bits of normal and fatty ultrasound liver images. The upper bits were compressed reversibly. This combination of techniques is termed PTSVQ with splitting. The effect of the compression on the difference in texture between normal and fatty liver images was studied at different compression rates and distortions. The changes in texture were measured by changes in the principal components of the covariance matrix of image vectors. The vectors were the same size as those used in the compression technique. There were clear differences in the components of normal and fatty liver images. These differences were largely removed by the PTSVQ with splitting technique even at average single pixel distortions several times smaller than the image noise. These results suggest that the effect of compression on second order statistics should be measured when evaluating algorithms in addition to the first order average distortion.