<title>Predictive tree-structured vector quantization for medical image compression and its evaluation with computerized image analysis</title>

Abstract

We present a predictive learning tree-structured vector quantization technique for medical image compression. A multi-layer perceptron (MLP) based vector predictor is employed to remove first as well as higher order correlations that exist among neighboring pixels. We use a learning tree-structured vector quantization (LTSVQ) scheme, which is 1)ased on competitive learning (CL) algorithm, to encode the residual vector. LTSVQ algorithm is computationally very efficient, easy to implement and provides performance comparable to that of LBG (Linde, Buzo and Gray) algorithm. We use computerized image analysis (image segmentation) as well as mean square error (MSE) and signal-to-noise ratio (SNR) to evaluate the quality of the compressed images. We apply the neural network based predictive LTSVQ to mammographic and magnetic resonance (MR) images, and evaluate the quality of images with different compression ratios.

Automatic detection of foreign objects in computed radiography

“…13 The automatic foreign object detection algorithm has been further integrated into our task-oriented image quality evaluation method which fully takes into account the clinical purpose of the medical images. 14 The task-oriented image quality evaluation method has been shown to be very promising in assessment of CR image quality for radiation dose optimization. 15 In this particular CR study, we can quantify the quality of those images acquired with different radiation doses by comparing the detected boundaries of the foreign objects and their segmented bone structures in addition to the wavelet analysis.…”

<title>Task-oriented lossy compression of magnetic resonance images</title>

“…The results of the semiautomatic segmentation using the ICM-based technique are shown inFigure 4.1 (b). Visually, the correspondence between the two classifications is easy to see: the automatic segmentation has classified most of the lesions identified by the radiologist, but some have been missed (false negatives), and some non-lesion tissue has been wrongly classified as lesion (false positives).Xuan et al[73] used the segmentation of compressed medical images as a method of quantifying the effects of compression loss on a typical medical image processing task. They compared the binary segmentations of original and compressed images, and used a percentage of misclassified pixels as the metric.…”