Dimitris Glotsos

Automated diagnosis of brain tumours astrocytomas using probabilistic neural network clustering and support vector machines

A computer-aided diagnosis system was developed for assisting brain astrocytomas malignancy grading. Microscopy images from 140 astrocytic biopsies were digitized and cell nuclei were automatically segmented using a Probabilistic Neural Network pixel-based clustering algorithm. A decision tree classification scheme was constructed to discriminate low, intermediate and high-grade tumours by analyzing nuclear features extracted from segmented nuclei with a Support Vector Machine classifier. Nuclei were segmented with an average accuracy of 86.5%. Low, intermediate, and high-grade tumours were identified with 95%, 88.3%, and 91% accuracies respectively. The proposed algorithm could be used as a second opinion tool for the histopathologists.

Robust Estimation of Bioaffinity Assay Fluorescence Signals

In this paper, the challenging problem of robust mean-signal estimation of a single-step microparticle bioaffinity assay is investigated. For this purpose, a density estimation-based robust algorithm (DER) was developed. The DER algorithm was comparatively evaluated with four other parameter estimation methods (mean value, median filtering, least square estimation, Welsch robust m-estimator). Two important questions were raised and investigated: 1) Which of the five methods can robustly estimate the mean bioaffinity signal? and 2) How many microparticles need to be measured in order to obtain an accurate estimate of the mean signal value? To answer the questions, bootstrap and coefficient of variation (CV) analyses were performed. In the CV analysis, the DER algorithm gave the best results: The CV ranged from 0.8% to 4.9% when the number of microparticles used for the mean signal estimation varied from 800 to 30. In the bootstrap analysis of the standard error, the DER algorithm had the smallest variance. As a conclusion, it can be underlined that: 1) of all methods tested, the DER algorithm gave the most consistent and reproducible results according to the bootstrap and CV analysis; 2) using the DER algorithm accurate estimates could be calculated based on 80-100 particles, corresponding to a typical assay measurement time of 1 min; and 3) the investigated bioaffinity signals contained a large number of outliers (observations that severely deviate from the majority of data) and therefore robust techniques were necessary for the mean signal estimation tasks

Pattern recognition system for the discrimination of multiple sclerosis from cerebral microangiopathy lesions based on texture analysis of magnetic resonance images.

In this study, a pattern recognition system has been developed for the discrimination of multiple sclerosis (MS) from cerebral microangiopathy (CM) lesions based on computer-assisted texture analysis of magnetic resonance images. Twenty-three textural features were calculated from MS and CM regions of interest, delineated by experienced radiologists on fluid attenuated inversion recovery images and obtained from 11 patients diagnosed with clinically definite MS and from 18 patients diagnosed with clinically definite CM. The probabilistic neural network classifier was used to construct the proposed pattern recognition system and the generalization of the system to unseen data was evaluated using an external cross validation process. According to the findings of the present study, statistically significant differences exist in the values of the textural features between CM and MS: MS regions were darker, of higher contrast, less homogeneous and rougher as compared to CM.

Complementary DNA Microarray Image Processing Based on the Fuzzy Gaussian Mixture Model

The objective of this paper was to investigate the segmentation ability of the fuzzy Gaussian mixture model (FGMM) clustering algorithm, applied on complementary DNA (cDNA) images. Following a standard established procedure, a simulated microarray image of 1600 cells, each containing one spot, was produced. For further evaluation of the algorithm, three real microarray images were also used, each containing 6400 spots. For the task of locating spot borders and surrounding background (BG) in each cell, an automatic gridding process was developed and applied on microarray images. The FGMM and the Gaussian mixture model (GMM) algorithms were applied to each cell with the purpose of discriminating foreground (FG) from BG. The segmentation abilities of both algorithms were evaluated by means of the segmentation matching factor, coefficient of determination, and concordance correlation, in respect to the actual classes (FG-BG pixels) of the simulated spots. Pairwise correlation and mean absolute error of the real images among replicates were also calculated. The FGMM was found to perform better and with equal processing time, as compared to the GMM, rendering the FGMM algorithm an efficient alternative for segmenting cDNA microarray images.

Breast Cancer Characterization Based on Image Classification of Tissue Sections Visualized under Low Magnification

Rapid assessment of tissue biopsies is a critical issue in modern histopathology. For breast cancer diagnosis, the shape of the nuclei and the architectural pattern of the tissue are evaluated under high and low magnifications, respectively. In this study, we focus on the development of a pattern classification system for the assessment of breast cancer images captured under low magnification (×10). Sixty-five regions of interest were selected from 60 images of breast cancer tissue sections. Texture analysis provided 30 textural features per image. Three different pattern recognition algorithms were employed (kNN, SVM, and PNN) for classifying the images into three malignancy grades: I–III. The classifiers were validated with leave-one-out (training) and cross-validation (testing) modes. The average discrimination efficiency of the kNN, SVM, and PNN classifiers in the training mode was close to 97%, 95%, and 97%, respectively, whereas in the test mode, the average classification accuracy achieved was 86%, 85%, and 90%, respectively. Assessment of breast cancer tissue sections could be applied in complex large-scale images using textural features and pattern classifiers. The proposed technique provides several benefits, such as speed of analysis and automation, and could potentially replace the laborious task of visual examination.

Improving accuracy in astrocytomas grading by integrating a robust least squares mapping driven support vector machine classifier into a two level grade classification scheme

Grading of astrocytomas is an important task for treatment planning; however, it suffers from significantly great inter-observer variability. Computer-assisted diagnosis systems have been propose to assist towards minimizing subjectivity, however, these systems present either moderate accuracy or utilize specialized staining protocols and grading systems that are difficult to apply in daily clinical practice. The present study proposes a robust mathematical formulation by integrating state-of-art technologies (support vector machines and least squares mapping) in a cascade classification scheme for separating low from high and grade III from grade IV astrocytic tumours. Results have indicated that low from high-grade tumours can be correctly separated with a certainty as high as 97.3%, whereas grade III from grade IV tumours with 97.8%. The overall performance was 95.2%. These high rates have been a result of applying the least squares mapping technique to features prior to classification. A significant byproduct of least squares mapping is that the number of support vectors of the SVM classifiers dropped dramatically from about 80% when no mapping was used to less than 5% when mapping was used. The latter is a clear indication that the SVM classifier has a greater potential to generalize well to new data. In this way, digital image analysis systems for automated grading of astrocytomas are brought closer to clinical practice.

An image-analysis system based on support vector machines for automatic grade diagnosis of brain-tumour astrocytomas in clinical routine

An image-analysis system based on the concept of Support Vector Machines (SVM) was developed to assist in grade diagnosis of brain tumour astrocytomas in clinical routine. One hundred and forty biopsies of astrocytomas were characterized according to the WHO system as grade II, III and IV. Images from biopsies were digitized, and cell nuclei regions were automatically detected by encoding texture variations in a set of wavelet, autocorrelation and parzen estimated descriptors and using an unsupervised SVM clustering methodology. Based on morphological and textural nuclear features, a decision-tree classification scheme distinguished between different grades of tumours employing an SVM classifier. The system was validated for clinical material collected from two different hospitals. On average, the SVM clustering algorithm correctly identified and accurately delineated 95% of all nuclei. Low-grade tumours were distinguished from high-grade tumours with an accuracy of 90.2% and grade III from grade IV with an accuracy of 88.3% The system was tested in a new clinical data set, and the classification rates were 87.5 and 83.8%, respectively. Segmentation and classification results are very encouraging, considering that the method was developed based on every-day clinical standards. The proposed methodology might be used in parallel with conventional grading to support the regular diagnostic procedure and reduce subjectivity in astrocytomas grading.

Segmentation of Complementary DNA Microarray Images by Wavelet-Based Markov Random Field Model

A wavelet-based modification of the Markov random field (WMRF) model is proposed for segmenting complementary DNA (cDNA) microarray images. For evaluation purposes, five simulated and a set of five real microarray images were used. The one-level stationary wavelet transform (SWT) of each microarray image was used to form two images, a denoised image, using hard thresholding filter, and a magnitude image, from the amplitudes of the horizontal and vertical components of SWT. Elements from these two images were suitably combined to form the WMRF model for segmenting spots from their background. The WMRF was compared against the conventional MRF and the Fuzzy C means (FCM) algorithms on simulated and real microarray images and their performances were evaluated by means of the segmentation matching factor (SMF) and the coefficient of determination (r 2). Additionally, the WMRF was compared against the SPOT and SCANALYZE, and performances were evaluated by the mean absolute error (MAE) and the coefficient of variation (CV). The WMRF performed more accurately than the MRF and FCM (SMF: 92.66, 92.15, and 89.22, r 2 : 0.92, 0.90, and 0.84, respectively) and achieved higher reproducibility than the MRF, SPOT, and SCANALYZE (MAE: 497, 1215, 1180, and 503, CV: 0.88, 1.15, 0.93, and 0.90, respectively).

A multi-classifier system for the characterization of normal, infectious, and cancerous prostate tissues employing transrectal ultrasound images

A computer-aided diagnostic system has been developed for the discrimination of normal, infectious and cancer prostate tissues based on texture analysis of transrectal ultrasound images. The proposed system has been designed using a panel of three classifiers, which have been evaluated individually or as a mutli-classifier scheme, using the external cross-validation procedure. Clinical data consisted of 165 transrectal ultrasound images, characterized by an experienced physician as normal (55/165), cancerous (55/165), and infectious (55/165) prostate cases. From each image, the physician delineated the most representative regions of interest, from which, 23 textural features were extracted. Classification was seen as a two level hierarchical decision tree. Normal from infectious and infectious from cancer cases were discriminated at the 1st and 2nd level of the decision tree, respectively. The best classification results for the 1st level were 89.5%, whereas for the 2nd level 90.1%. The utilization of multi-classifier system improved the discrimination of prostate pathologies as compared to individual classifiers; for infectious prostate cases improvement was from 87.3% to 88.7% and for cancer prostate cases improvement was from 84.1% to 91.4%. In terms of overall system performance (the decision trees node propagating error taken into account), best classification accuracies were 89.5%, 79.6% and 82.7% for the recognition of normal, infectious and cancer cases, respectively. The proposed system might be used as a second opinion tool for assisting diagnosis of different prostate pathologies.

A wavelet-based Markov random field segmentation model in segmenting microarray experiments

In the present study, an adaptation of the Markov Random Field (MRF) segmentation model, by means of the stationary wavelet transform (SWT), applied to complementary DNA (cDNA) microarray images is proposed (WMRF). A 3-level decomposition scheme of the initial microarray image was performed, followed by a soft thresholding filtering technique. With the inverse process, a Denoised image was created. In addition, by using the Amplitudes of the filtered wavelet Horizontal and Vertical images at each level, three different Magnitudes were formed. These images were combined with the Denoised one to create the proposed SMRF segmentation model. For numerical evaluation of the segmentation accuracy, the segmentation matching factor (SMF), the Coefficient of Determination (r(2)), and the concordance correlation (p(c)) were calculated on the simulated images. In addition, the SMRF performance was contrasted to the Fuzzy C Means (FCM), Gaussian Mixture Models (GMM), Fuzzy GMM (FGMM), and the conventional MRF techniques. Indirect accuracy performances were also tested on the experimental images by means of the Mean Absolute Error (MAE) and the Coefficient of Variation (CV). In the latter case, SPOT and SCANALYZE software results were also tested. In the former case, SMRF attained the best SMF, r(2), and p(c) (92.66%, 0.923, and 0.88, respectively) scores, whereas, in the latter case scored MAE and CV, 497 and 0.88, respectively. The results and support the performance superiority of the SMRF algorithm in segmenting cDNA images.