Antonis Daskalakis

Improving brain tumor characterization on MRI by probabilistic neural networks and non-linear transformation of textural features

The aim of the present study was to design, implement and evaluate a software system for discriminating between metastatic and primary brain tumors (gliomas and meningiomas) on MRI, employing textural features from routinely taken T1 post-contrast images. The proposed classifier is a modified probabilistic neural network (PNN), incorporating a non-linear least squares features transformation (LSFT) into the PNN classifier. Thirty-six textural features were extracted from each one of 67 T1-weighted post-contrast MR images (21 metastases, 19 meningiomas and 27 gliomas). LSFT enhanced the performance of the PNN, achieving classification accuracies of 95.24% for discriminating between metastatic and primary tumors and 93.48% for distinguishing gliomas from meningiomas. To improve the generalization of the proposed classification system, the external cross-validation method was also used, resulting in 71.43% and 81.25% accuracies in distinguishing metastatic from primary tumors and gliomas from meningiomas, respectively. LSFT improved PNN performance, increased class separability and resulted in dimensionality reduction.

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.

Computerized Analysis of Digital Subtraction Angiography: A Tool for Quantitative In-vivo Vascular Imaging

The purpose of our study was to develop a user-independent computerized tool for the automated segmentation and quantitative assessment of in vivo-acquired digital subtraction angiography (DSA) images. Vessel enhancement was accomplished based on the concept of image structural tensor. The developed software was tested on a series of DSA images acquired from one animal and two human angiogenesis models. Its performance was evaluated against manually segmented images. A receiver’s operating characteristic curve was obtained for every image with regard to the different percentages of the image histogram. The area under the mean curve was 0.89 for the experimental angiogenesis model and 0.76 and 0.86 for the two clinical angiogenesis models. The coordinates of the operating point were 8.3% false positive rate and 92.8% true positive rate for the experimental model. Correspondingly for clinical angiogenesis models, the coordinates were 8.6% false positive rate and 89.2% true positive rate and 9.8% false positive rate and 93.8% true positive rate, respectively. A new user-friendly tool for the analysis of vascular networks in DSA images was developed that can be easily used in either experimental or clinical studies. Its main characteristics are robustness and fast and automatic execution.

Non-linear least squares features transformation for improving the performance of probabilistic neural networks in classifying human brain tumors on MRI

The aim of the present study was to design, implement, and evaluate a software system for discriminating between metastases, meningiomas, and gliomas on MRI. The proposed classifier is a modified probabilistic neural network (PNN), incorporating a second degree least squares features transformation (LSFT) into the PNN classifier. Thirty-six textural features were extracted from each one of 75 T1-weighted post-contrast MR images (24 metastases, 21 meningiomas, and 30 gliomas). Classification performance was evaluated employing the leave-one-out method and for all possible textural feature combinations. LSFT enhanced the performance of the PNN, achieving 93.33%in discriminating between the three major types of human brain tumors, against 89.33% scored by the PNN alone. Best feature combination for achieving highest discrimination power included the mean value and entropy, which reflect specific properties of texture, i.e. signal strength and inhomogeneity. LSFT improved PNN performance, increased class separability, and resulted in dimensionality reduction.

Colour-Texture based image analysis method for assessing the Hormone Receptors status in Breast tissue sections

Hormone receptors have been used in prognosis of breast carcinomas and their positive status is of clinical value in hormonal therapy. Determination of this status is based on the subjective visual inspection of the stained nuclei in the specimens. The aim of this study was the assessment of the estrogen receptors (ER) positive status of breast carcinomas, by means of colour-texture based image analysis methodology. Twenty two cases of immunohistochemically (IHC) stained breast biopsies were initially assessed by a histopathologist for ER positive status, following a clinical scoring protocol. Custom-designed image analysis software was developed for automatically assessing the ER positive status, employing colour textural features and the k-Nearest Neighbor weighted votes classification algorithm. Computer-based image analysis system resulted in 86.4% overall accuracy and in 0.875 Kendalls coefficient of concordance (p<0.001), ranking correctly 19/22 cases. Colour-texture analysis of IHC stained specimens might have an impact in the quantitative assessment of ER status.

Assessing estrogen receptors' status by texture analysis of breast tissue specimens and pattern recognition methods

An image analysis system (IAS) was developed for the quantitative assessment of estrogen receptors (ER) positive status from breast tissue microscopy images. Twenty-four cases of breast cancer biopsies, immunohisto-chemically (IHC) stained for ER, were microscopically assessed by a histopathologist, following a clinical routine scoring protocol. Digitized microscopy views of the specimens were used in the IASs design. IAS comprised a/image segmentation, for nuclei determination, b/extraction of textural features, by processing of nuclei-images utilizing the Laws and Gabor filters and by calculating textural features from the processed nuclei-images, and c/PNN and SVM classifiers design, for discriminating positively stained nuclei. The proportion of the latter in each cases images was compared against the physicians score. Using Spearmans rank correlation, high correlation was found between the histo-pathogists and IASs scores (rho=0.89, p<0.001) and 22/24 cases were correctly characterised, indicating IASs reliability in the quantitative evaluation of ER as additional assistance to physicians assessment.

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.

Technical Section: A hybrid pixel-based classification method for blood vessel segmentation and aneurysm detection on CTA

In the present study, a hybrid semi-supervised pixel-based classification algorithm is proposed for the automatic segmentation of intracranial aneurysms in Computed Tomography Angiography images. The algorithm was designed to discriminate image pixels as belonging to one of the two classes: blood vessel and brain parenchyma. Its accuracy in vessel and aneurysm detection was compared with two other reliable methods that have already been applied in vessel segmentation applications: (a) an advanced and novel thresholding technique, namely the frequency histogram of connected elements (FHCE), and (b) the gradient vector flow snake. The comparison was performed by means of the segmentation matching factor (SMF) that expressed how precise and reproducible was the vessel and aneurysm segmentation result of each method against the manual segmentation of an experienced radiologist, who was considered as the gold standard. Results showed a superior SMF for the hybrid (SMF=88.4%) and snake (SMF=87.2%) methods compared to the FHCE (SMF=68.9%). The major advantage of the proposed hybrid method is that it requires no a priori knowledge of the topology of the vessels and no operator intervention, in contrast to the other methods examined. The hybrid method was efficient enough for use in 3D blood vessel reconstruction.

A comparative study of individual and ensemble majority vote cDNA microarray image segmentation schemes, originating from a spot-adjustable based restoration framework

The aim of this study was to comparatively evaluate the performances of various segmentation algorithms, in conjunction with a noise reduction step, for gene expression levels intensity extraction in cDNA microarray images. Different segmentation algorithms, based on histogram and unsupervised classification methods, which have never been previously employed in microarray image analysis, were employed either individually or in ensemble majority vote structures for separating spot-images from background pixels. The performances of segmentation algorithms or ensemble structures were evaluated by assessing the validity and reproducibility of gene expression levels extraction in simulated and real cDNA microarray images. By processing high quality simulated images, the highest segmentation accuracy was achieved by an ensemble structure (Histogram Concavity, Gaussian Kernelized Fuzzy-C-Means, Seeded Region Growing). Optimum performance in terms of processing time and segmentation precision for low quality simulated and replicated real cDNA microarray images was attained by the Histogram Concavity algorithm.

PDA-based system with teleradiology and image analysis capabilities

The aim of the present study was to design and implement a Personal Digital Assistant (PDA)-based teleradiology system incorporating image processing and analysis facilities for use in emergency situations within a hospital environment. The system comprised a DICOM-server, connected to an MRI unit, 3 wireless access points, and 3 PDAs (HP iPaq rx3715). PDA application software was developed in MS Embedded Visual C++ 4.0. Each PDA can receive, load, process and analyze hi-quality static MR images. Image processing includes gray-scale manipulation and spatial filtering techniques while image analysis incorporates a probabilistic neural network (PNN) classifier, which was optimally designed employing a suitable combination of textural features and was evaluated using the leave-one-out method. The PNN is capable of discriminating between three major types of human brain tumors with accuracy of 86.66%. The developed application may be useful as a mobile medical teleconsultation tool.