April Khademi

Pan-Canadian Evaluation of Irreversible Compression Ratios (“Lossy” Compression) for Development of National Guidelines

New technological advancements including multislice CT scanners and functional MRI, have dramatically increased the size and number of digital images generated by medical imaging departments. Despite the fact that the cost of storage is dropping, the savings are largely surpassed by the increasing volume of data being generated. While local area network bandwidth within a hospital is adequate for timely access to imaging data, efficiently moving the data between institutions requires wide area network bandwidth, which has a limited availability at a national level. A solution to address those issues is the use of lossy compression as long as there is no loss of relevant information. The goal of this study was to determine levels at which lossy compression can be confidently used in diagnostic imaging applications. In order to provide a fair assessment of existing compression tools, we tested and compared the two most commonly adopted DISCOM compression algorithms: JPEG and JPEG-2000. We conducted an extensive pan-Canadian evaluation of lossy compression applied to seven anatomical areas and five modalities using two recognized techniques: objective methods or diagnostic accuracy and subjective assessment based on Just Noticeable Difference. By incorporating both diagnostic accuracy and subjective evaluation techniques, enabled us to define a range of compression for each modality and body part tested. The results of our study suggest that at low levels of compression, there was no significant difference between the performance of lossy JPEG and lossy JPEG 2000, and that they are both appropriate to use for reporting on medical images. At higher levels, lossy JPEG proved to be more effective than JPEG 2000 in some cases, mainly neuro CT. More evaluation is required to assess the effect of compression on thin slice CT. We provide a table of recommended compression ratios for each modality and anatomical area investigated, to be integrated in the Canadian Association of Radiologists standard for the use of lossy compression in medical imaging.

Robust White Matter Lesion Segmentation in FLAIR MRI

This paper discusses a white matter lesion (WML) segmentation scheme for fluid attenuation inversion recovery (FLAIR) MRI. The method computes the volume of lesions with subvoxel precision by accounting for the partial volume averaging (PVA) artifact. As WMLs are related to stroke and carotid disease, accurate volume measurements are most important. Manual volume computation is laborious, subjective, time consuming, and error prone. Automated methods are a nice alternative since they quantify WML volumes in an objective, efficient, and reliable manner. PVA is initially modeled with a localized edge strength measure since PVA resides in the boundaries between tissues. This map is computed in 3-D and is transformed to a global representation to increase robustness to noise. Significant edges correspond to PVA voxels, which are used to find the PVA fraction α (amount of each tissue present in mixture voxels). Results on simulated and real FLAIR images show high WML segmentation performance compared to ground truth (98.9% and 83% overlap, respectively), which outperforms other methods. Lesion load studies are included that automatically analyze WML volumes for each brain hemisphere separately. This technique does not require any distributional assumptions/parameters or training samples and is applied on a single MR modality, which is a major advantage compared to the traditional methods.

Shift-invariant discrete wavelet transform analysis for retinal image classification

This work involves retinal image classification and a novel analysis system was developed. From the compressed domain, the proposed scheme extracts textural features from wavelet coefficients, which describe the relative homogeneity of localized areas of the retinal images. Since the discrete wavelet transform (DWT) is shift-variant, a shift-invariant DWT was explored to ensure that a robust feature set was extracted. To combat the small database size, linear discriminant analysis classification was used with the leave one out method. 38 normal and 48 abnormal (exudates, large drusens, fine drusens, choroidal neovascularization, central vein and artery occlusion, histoplasmosis, arteriosclerotic retinopathy, hemi-central retinal vein occlusion and more) were used and a specificity of 79% and sensitivity of 85.4% were achieved (the average classification rate is 82.2%). The success of the system can be accounted to the highly robust feature set which included translation, scale and semi-rotational, features. Additionally, this technique is database independent since the features were specifically tuned to the pathologies of the human eye.

Automatic contrast enhancement of white matter lesions in FLAIR MRI

This work concerns the development of a novel contrast enhancement algorithm for FLAIR-weighted cerebral MRI with white matter lesions (WML). The proposed method utilizes both a robust estimate of edge magnitude and intensity values to discriminate between pathological and non-pathological information. These two features are combined through several transformations, such that WML are highlighted, and normal appearing white/gray matter are suppressed. The technique utilizes information solely computed from each image and thus adapts to the input images characteristics. The results show a significant improvement of the contrast between white matter lesions and other brain tissue (average contrast improvement of 41.1%). To demonstrate the robustness of such an enhancement scheme for WML analysis, a threshold-based segmenter is applied, which extracts the WML with good results.

Small bowel image classification using cross-co-occurrence matrices on wavelet domain

This paper presents a novel system to compute the automated classification of wireless capsule endoscope images. Classification is achieved by a classical statistical approach, but novel features are extracted from the wavelet domain and they contain both color and texture information. First, a shift-invariant discrete wavelet transform (SIDWT) is computed to ensure that the multiresolution feature extraction scheme is robust to shifts. The SIDWT expands the signal (in a shift-invariant way) over the basis functions which maximize information. Then cross-co-occurrence matrices of wavelet subbands are calculated and used to extract both texture and color information. Canonical discriminant analysis is utilized to reduce the feature space and then a simple 1D classifier with the leave one out method is used to automatically classify normal and abnormal small bowel images. A classification rate of 94.7% is achieved with a database of 75 images (41 normal and 34 abnormal cases). The high success rate could be attributed to the robust feature set which combines multiresolutional color and texture features, with shift, scale and semi-rotational invariance. This result is very promising and the method could be used in a computer-aided diagnosis system or a content-based image retrieval scheme.

Medical image texture analysis: A case study with small bowel, retinal and mammogram images

This work concerns the development of a generalized framework for computer-aided diagnosis of medical images. The system is built to mimic human texture perception as texture has been shown to be an important feature for pathology discrimination in medical images. In particular, it was shown by Julesz that orientation, frequency and scale are important markers for texture discrimination. Consequently, this work focuses on the design of a feature extraction scheme which identifies these texture markers (in accordance to Juleszpsilas human texture perception model). To get a rich description of the space-localized texture elements, wavelet analysis is employed using a scale-invariant representation. A robust, multiscale texture analysis scheme is employed to quantify the texture characteristics of the image. Wavelet-domain graylevel cooccurrence matrices were implemented in a variety of directions in order to capture the orientation of such texture elements (which also offered semi-rotational invariance). To test the systempsilas performance, retinal, small bowel and mammogram images were used. 75 small bowel images were correctly classified at an average classification accuracy of 85%, 86 retinal images had an average classification accuracy of 82.2% and the mammogram lesions (54) were classified correctly 69% on average.

Comparison of JPEG 2000 and Other Lossless Compression Schemes for Digital Mammograms

In this study, we propose JPEG 2000 as an algorithm for the compression of digital mammograms and the proposed work is the first real-time implementation of JPEG 2000 on a mammogram image database. Only the lossless compression mode of JPEG 2000 was examined to ensure that the mammogram is delivered without distortion. The performance of JPEG 2000 was compared against several other lossless coders: JPEG-LS, lossless-JPEG, adaptive Huffman, arithmetic with a zero order and a first order probability model and Lempel-Ziv Welch (EZW) with 12 and 15 bit dictionaries. Each compressor was supplied the identical set of 50 mammograms, each having a resolution of 8bits/pixel and dimensions of 1024times1024. Experimental results indicate JPEG 2000 and JPEG-LS provide comparable compression performance since their compression ratios differed by 0.72% and both compressors also superseded the results of the other coders. Although JPEG 2000 suffered from a slightly longer encoding and decoding delay than JPEG-ES (0.8 s on average), it is still preferred for mammogram images due to the wide variety of features that aid in reliable image transmission, provide an efficient mechanism for remote access to digital libraries and contribute to fast database access

Nonparametric statistical tests for exploration of correlation and nonstationarity in images

This work proposes two statistical-based techniques to quantify (with confidence) whether random 2D data (images) are correlated or nonstationary. Traditionally, such exploratory data analysis techniques have been developed for 1D signals, such as EEG. This paper presents a new application of Mantels test for clustering to examine spatial dependence and a novel 2D extension of the traditional 1D version of the reverse arrangements test to examine data nonstationary. Simulated data (correlated and nonstationary) were generated and subject to several rotations, scales and translations, in order to test the robustness of the techniques. Mantels test for clustering correctly classified the images as correlated for 100% of the cases (including those with rotations, scales and translations (RSTs)). For the 2D extension of the reverse arrangements test, the linear trend analysis correctly found 15/16 regions to have pixel-wise nonstationarity, and the nonlinear trend analysis correctly classified nonstationarity in all but two cases (14/16) (for all RSTs). As a result of the high classification rates, the techniques are relatively invariant to changes in RST. These two statistical tests have a variety of applications in medical imaging (i.e. modeling), and are discussed in this work. An additional application of the work is presented in the end, demonstrating the possibility that such test statistics may be used as features to classify different textures.

Image Enhancement and Noise Suppression for FLAIR MRIs With White Matter Lesions

This work presents an image reconstruction technique for noise suppression in FLAIR MRI with white matter lesions (WML). The technique utilizes a fuzzy edge estimate to initially localize edge information. Edge and intensity information are coupled through the conditional expectation operator, resulting in a robust and global description of the edge content in the image. As this global measure separates noise and useful edge information, a threshold is used to suppress the irrelevant details and integration is used to reconstruct the “noisefree” image. The threshold is automatically determined based on an objective function that minimizes noise (within class scatter) while maximizing contrast between the WML and brain tissue classes (between class variance). The result is an edge preserving smoothing filter, since the image is reconstructed based on the edge map. The proposed method was compared to the bilateral filter and was found to provide on average a 44.39% increase of noise attenuation in flat regions (smoothness) and a 34.14% increase in edge amplification (enhancement).

Edge-based partial volume averaging estimation for FLAIR MRI with white matter lesions

Through the combination of intensity and fuzzy edge strength measures, a new partial volume averaging (PVA) quantification technique for FLAIR MRI with white matter lesions (WML) is developed. It is focused on an edge-based approach, which “probes” for PVA voxels via a global estimate for the change in the proportion of tissues α′. This estimate is refined according to a probabilistic threshold, and the result is decoded to find the proportion of tissues fraction α - the percentage of one tissue found in a mixture voxel. The results from several images are shown illustrating how the technique may be used to segment PVA and pure tissue classes. The result is a non-model based approach to the detection and quantification of PVA.