Mammography has a central role in screening and diagnosis of breast lesions, allowing early detection of the pathology and reduction of fatal cases. Deep Convolutional Neural Networks have shown a great potentiality to address the issue of early detection of breast cancer with an acceptable level of accuracy and reproducibility. In the present paper, we illustrate the development of a deep learning study aimed to process and classify lesions in mammograms with the use of slender neural networks not yet used in literature. For this reason, a traditional convolution network was compared with a novel one obtained making use of much more efficient depth wise separable convolution layers. Preliminary numerical results are detailed and future plans outlined.

This work aimed to investigate whether automated classifiers belonging to feature-based and deep learning may approach brain metastases segmentation successfully. Support Vector Machine and V-Net Convolutional Neural Network are selected as representatives of the two approaches. In the experiments, we consider several configurations of the two methods to segment brain metastases on contrast-enhanced T1-weighted magnetic resonance images. Performances were evaluated and compared under critical conditions imposed by the clinical radiotherapy domain, using in-house dataset and public dataset created for the Multimodal Brain Tumour Image Segmentation (BraTS) challenge. Our results showed that the feature-based and the deep network approaches are promising for the segmentation of Magnetic Resonance Imaging (MRI) brain metastases achieving both an acceptable level of performance. Experimental results also highlight different behaviour between the two methods. Support vector machine (SVM) improves performance with a smaller training set, but it is unable to manage a high level of heterogeneity in the data and requires post-processing refinement stages. The V-Net model shows good performances when trained on multiple heterogeneous cases but requires data augmentations and transfer learning procedures to optimise its behaviour. The paper illustrates a software package implementing an integrated set of procedures for active support in segmenting brain metastases within the radiotherapy workflow.

In computational neuroimaging, the analysis of functional Magnetic Resonance Images (fMRIs) using fuzzy clustering methods is a promising data driven approach to explore brain functional connectivity. In this complex domain, accurate evaluation procedures based on suitable indexes, able to identify optimal clustering results, are of great values strongly affecting the validity and interpretation of the overall fMRI data analysis. A large number of clustering validation indexes have been proposed in literature. This work proposes a comparison analysis of eight representative fuzzy and crisp clustering validation indexes. Salient aspects of the proposed strategy are the use of the widely adopted fuzzy c-means algorithm as underlying fuzzy clustering algorithm and the use of resting state fMRI data from the NITRC repository.

RS-fMRI data analysis for functional connectivity explorations is a challenging topic in computational neuroimaging. Several approaches have been investigated to discover whole-brain data features. Among these, clustering techniques based on Competitive Learning (CL) and Spectral Methods (SM) have been shown effective in providing useful information in various contexts. We selected three clustering algorithms and two spectral methods, i.e the clustering algorithm are Self-organising Maps (SOM), Neural Gas (NG) and Growing Neural Gas (GNG), whereas the spectral methods are the classic Principal Component Analysis (PCA) and the Nonlinear Robust Fuzzy Principal Component Analysis (NRFPCA). We validated clustering with Davies–Bouldin Index (DBI) and we selected informative principal components using Random Matrix Theory (RMT). tools. We adopted these techniques to study the intrinsic functional properties of images coming from a shared repository of resting state fMRI experiments (1000 Functional Connectome Project).

We used model-free methods to explore the brain’s functional properties adopting a partitioning procedure based on cross-clustering. We selected Fuzzy C-Means (FCM) and Neural Gas (NG) algorithms to find spatial patterns with temporal features and temporal patterns with spatial features. We applied these algorithms to a shared fMRI repository of face recognition tasks. We matched the classes found and our results of functional connectivity analysis with partitioning of BOLD signal signatures. We compared the outcomes using the just acquired model-based knowledge as likely ground truth, confirming the role of Fusiform Brain Regions. In general, partitioning results show a better spatial clustering than temporal clustering for both algorithms. In the case of temporal clustering, FCM outperforms Neural Gas. The relevance of brain subregions related to face recognition were correctly distinguished by the algorithms and the results are in agreement with the current neuroscientific literature.

This paper presents a fuzzy logic framework for dental caries and erosion risk assessment. Two interdependent modules are implemented within a cloud architecture. The first module is a fuzzy expert system designed for physicians and expert users, able to provide an active support in formulating risk judgements. The second module is oriented to generic users for oral health promotion. Conceptual ingredients of the fuzzy logic framework are principally defined by eliciting knowledge from a group of experts. The generation of rules involves both structured interviews and data driven learning procedures based on the use of neuro-fuzzy techniques.

The objective of this study is to develop a semi-automatic, interactive segmentation strategy for efficient and accurate brain metastases delineation on Post Gadolinium T1-weighted brain MRI images. Salient aspects of the proposed solutions are the combined use of machine learning and image processing techniques, based on Support Vector Machine and Morphological Operators respectively, to delineate pathological and healthy tissues. The overall segmentation procedure is designed to operate on a clinical setting to reduce the workload of health-care professionals but leaving to them full control of the process. The segmentation process was validated for in-house collected image data obtained from radiation therapy studies. The results prove that the allied use of SVM and Morphological Operators produces accurate segmentations, useful for their insertion in clinical practice.

In the present work, we investigate the usefulness of a new representation of the results obtained by fMRI data analysis, named weighted activation vector (WAV), built based on statistical parametric mapping. A software package for the generation and management of WAVs is illustrated. It is designed to support single-subject, multi-temporal and collective brain tumour studies. As seen in our experimental context, the combined use of WAVs and statistical parametric maps (SPMs) improves the quality of medical decisions before and after neurosurgical practice. Clustering techniques applied to WAVs can be efficiently analysed and optimised in an attempt to discover relevant properties of collective data.

This work focuses the attention on the segmentation of meningioma and peritumoral edema from multispectral brain MR imagery. Precise tumour and edema delineation and volume quantification from preoperative MRI data contribute to formulate surgical indications in elderly patients harbouring intracranial meningioma. The authors propose a fully automatic procedure based on the allied use of Graph Cut and support vector machine. The overall strategy combines the advantages of the image-based and machine learning techniques adopted, optimising the balancing between accuracy and stability/reproducibility of the results. Experimental results, obtained by processing in-house collected data, prove that the method is robust and oriented to the use in clinical practice.

RS-fMRI data analysis for functional connectivity explorations is a challenging topic in computational neuroimaging. Several approaches have been investigated to discover whole-brain data features. Among these, clustering techniques based on Soft Competitive Learning (SCL) have been shown effective in providing useful information in various contexts. However, although significant achievements have been reached, these techniques still present critical aspects that require further investigations. We selected three clustering algorithms, i.e. Self-Organizing Maps (SOM), Neural Gas (NG) and Growing Neural Gas (GNG), to study the intrinsic functional properties of images coming from a shared repository of resting state fMRI experiments (1000 Functional Connectome Project, i.e. Oxford dataset). To compare the functional connectivity based on soft clustering, we calculated the Seed Based Linear Correlation (SBLC) to study the Default Mode Network (DMN) functionality, i.e. we found that Precuneus L/R has the higher Correlations Coefficients with its controlateral part and with the posterior division of Cingulate Gyrus. The differences among the three soft clustering algorithms adopted were measured basing on Jaccard Similarity Coefficient (JSC), whereas the quality of clusters has been evaluated with Davies-Bouldin Index (DBI). The optimal clustering computation was with 2 partitions for all the algorithms. We obtained the following results: a) clusters differentiated the amplitude of BOLD signals for both Males and Females, i.e. low level signal vs high level signal; b) clusters also differentitated the quality of seedbased correlations, i.e. strong (positive) associations vs weakly associations. These multivariate outcomes highlighted the complementarty usage of clustering algorithms with statistical signal processing: the first made the partions, the last explain the partions. Alberto A. Vergani, Elisabetta Binaghi, Samuele Martinelli, Sabina Strocchi

We present a soft separation measure to validate fuzzy clustering results without defuzzyficaton. It is the generalization of Davies-Bouldin validation index (DB) for crisp clustering in the soft clustering domain; we named the measure Soft Davies-Bouldin index (SDB). We compared DB and SDB when applied to k-means and fuzzy c-means algorithms using eight datasets with ground-truth and two experimental fMRI datasets without ground-truth. We found that i) in more than half datasets, the optimal score of Soft Davies-Bouldin index was less than Davies-Bouldin index, ii) in half datasets that have ground-truth, the optimal score of Soft Davies-Bouldin index was less than Davies-Bouldin index in correspondence of the truth number of patterns, iii) the Soft Davies-Bouldin index outperformed the Davies-Bouldin index as central tendency of all datasets along the complete range of clusters considered.

This paper proposes a new accuracy evaluation method within a behavioral comparison strategy which uses interval type-2 fuzzy sets and derived operations to model reference data and define soft accuracy indexes. The method addresses the case in which grades of membership, collected by surveying experts, will often be different for the same reference pattern, because the experts will not necessarily be in agreement. The approach is illustrated using simple examples and an application in the domain of biomedical image segmentation.

In the present work we use pattern vectors derived from Statistical Parametric Map, generated from a group of artificial and in-house collected fMRI data, to conduct cluster analysis. Two clustering algorithms, self-organizing map (SOM) and growing neural gas (GNG), are selected to explore inherent properties in the brain functional data. As seen in our experimental context, SOM and GNG show comparable behavior, however GNG prevails in the management of large data sets. An exploratory, descriptive analysis is conducted on in-house collected data clustered by GNG and results are detailed in the paper.

This work quantitatively evaluates the effects induced by susceptibility characteristics of materials commonly used in dental practice on the quality of head MR images in a clinical 1.5T device. The proposed evaluation procedure measures the image artifacts induced by susceptibility in MR images by providing an index consistent with the global degradation as perceived by the experts. Susceptibility artifacts were evaluated in a near-clinical setup, using a phantom with susceptibility and geometric characteristics similar to that of a human head. We tested different dentist materials, called PAL Keramit, Ti6Al4V-ELI, Keramit NP, ILOR F, Zirconia and used different clinical MR acquisition sequences, such as “classical” SE and fast, gradient, and diffusion sequences. The evaluation is designed as a matching process between reference and artifacts affected images recording the same scene. The extent of the degradation induced by susceptibility is then measured in terms of similarity with the corresponding reference image. The matching process involves a multimodal registration task and the use an adequate similarity index psychophysically validated, based on correlation coefficient. The proposed analyses are integrated within a computer-supported procedure that interactively guides the users in the different phases of the evaluation method. 2-Dimensional and 3-dimensional indexes are used for each material and each acquisition sequence. From these, we drew a ranking of the materials, averaging the results obtained. Zirconia and ILOR F appear to be the best choice from the susceptibility artefacts point of view, followed, in order, by PAL Keramit, Ti6Al4V-ELI and Keramit NP.

In this work, we propose a novel behavioural comparison strategy specifically oriented to accuracy assessment in MRI glial tumour segmentation studies. A salient aspect of the proposed strategy is the use of the fuzzy set framework in modelling visual inspection and interpretation processes. In particular, a reference estimation strategy based on fuzzy connectedness principles is designed to merge individual labels and produce a common segmentation. The estimation is based exclusively on highly reliable partial information provided by experts. Interaction is then drastically limited compared with a complete manual tracing, leaving the estimation of the complete segmentation to the fuzzy connectedness method. A set of experiments was conceived and conducted to evaluate the contribution of the solutions proposed in the process of truth label collection and reference data estimation. A comparison analysis was also developed to see whether our method could constitute a worthy alternative to well-known and state-of-the-art solutions.

Event Abstract Back to Event FSL-BASED HYBRID ATLAS PROMOTES ACTIVATION WEIGHTED VECTOR ANALYSIS IN FUNCTIONAL NEURORADIOLOGY Alberto A. Vergani1*, Renzo Minotto2, Sabina Strocchi3 and Elisabetta Binaghi1 1 University of Insubria, Department of Theoretical and Applied Science, Italy 2 Ospedale di Circolo e Fondazione Macchi, O.U. Neuroradiology, Italy 3 Ospedale di Circolo e Fondazione Macchi, O.U. Health Physics, Italy Introduction and Motivations. Functional magnetic resonance methods detect in vivo brain hemodynamic responses related to specific task. After defined statistical processes pipeline, relevant information contained in the fMRI Blood-Oxygen-Level-Dependent signals are transformed in a 3D Statistical Parametric Maps. In our previous works, we investigated how to conveniently summarize these original 3D distributions by a more compact description composed of a set of indexes calculated for each functional area. These values, properly named Activation Weighted Indexes (AWIs), vary between 0 and 1 and regard all brain regions contained into specific template used for co-registration (e.g. using Juelich standard there will be 121 indexes). The proposed AWI based feature extraction procedure reduces the three-dimensions SPM distribution to the one-dimension AWI distribution The resulting quantized data structure of task-related brain activations, that we named Activation Weighted Vector (AWV), is a histogram, where each bin represents the level of the functionality of each brain structure. The expressiveness of AWV depends by the normalization procedure for fMRI scans: the more regions standard atlas has, the more AWIs are computable. Proceeding from these considerations, we addressed the problem of generating an inclusive template for AWI analysis satisfying the right trade-off between exhaustion power and visual usability. Description of CRAIIM Hybrid Atlas. The hybrid atlas is the outcome of the joint operation between different brain templates found into the FMRIB Software Library. They are Juelich histological atlas and Harvard-Oxford cortical and subcortical structural atlases. Both are probabilistic labelled and registered in MNI152 space. Juelich model was created by averaging multi-subject post-mortem cyto and myelo-architectonic segmentations, which has detected 52 grey matter structures and 10 white matters ones. Harvard-Oxford models cover 48 cortical and 21 subcortical structural regions, computed by segmentation of T1-weighted images of healthy male and female. The joint process was possible choosing a reference atlas and then adding lacking anatomical structures. In our case, Juelich atlas was the template from which regions dearth were easily included choosing them from Harvard-Oxford atlases. This procedure, shaped with complete or partial union operations, gives rise to an hybrid atlas that covers 161 regions, in which 121 are the 100% of Juelich, and other 40 are a variable percentage of Harvard-Oxford original brain volumes. The benefits of this hybrid atlas are the integration of fundamental neuroanatomy models useful for co-registration that in the standard template were absent, e.g. many frontal and temporal cortexes, subcallosal portions, cingulate gyrus and thalamus halves. The limitation is that these last regions are in some cases a minor proportion of the Harvard-Oxford template. The hybridized atlas is in NIFTI format and its FSL-like name is CRAIIM-thr0-1mm.nii.gz (where thr0 is a voxel probability threshold, i.e. Prob. ≥ 0 to belong a certain anatomical label; and 1mm means the voxels resolution). Clinical Applications. As above described, CRAIIM hybrid atlas has 161 regions. With this registration template, the AWV procedure generates vectors of 161 indexes from each SPMs yielding important advantages both in the analysis of collection of data and in the individual data interpretation. The data represented in a “well dimensioned” vector space are used to perform efficiently quantitative measures of dissimilarity among brains functional activations basing on a given selected metrics. Unsupervised learning methods are naturally applied to learn the structure of the data and then to discover and recognize salient neurological patterns that unearth hidden features, also permitting their interpretation like individual brain signatures. With AWV analysis, the Radiologists investigate activation about macro categories alike white/grey matters contribution or left/right hemispheres functional balancement. In addition, AWVs allow physician to examine geographically defined zones such as frontal or temporal cortexes, motor or visual systems, limbic core, callosum commissure, or more detailed contribution akin Broca’s areas, Wernicke’s areas, &c. AWVs also facilitate severe assessment modalities for patients monitoring. For example, within homogenous group, clinicians could discover interesting level differences; or between heterogeneous ones, they could observe interesting likeness. In addition, including time factor for each subject, longitudinal studies based on AWV could highlight brain plasticity processes more suitable than classical SPM qualitative visual inspection. These kind of clinical evaluations have claim among pre/post neurosurgical operation controls or during long-term ordinary patients’ checks. Obviously, AWV analysis is a math tool well applicable with active paradigm as well as passive ones, e.g. for resting state fMRI acquisition, keeping in mind the greater role that could have distributed functional connectivity instead locally defined expected activation. Present and Future Works. Hybrid atlas has enhanced activated weighted vector analysis thanks to its peculiarities. It has clinical utility for AWV methodology, but some kind of incompleteness for all brain region representation: it is missing of special structure like cerebellum and, in general, has some portion that is a percentage of its original atlases. Other own quality regards the resolution about only 1mm. Future work is planned to generalize the atlas building procedure with which to easily generate solutions for diversified functional neuroradiological applications honouring the salience of existing specialized template. Acknowledgements The authors would like to thank all staff of CRAIIM (Research Centre in Image Analysis and Medical Informatics). In particular, a special gratitude for Dr. Valentina Pedoia for her technical support and Dr. Sergio Balbi for his clinical suggestions. This work will be presented at the INCF Neuroinformatics Conference 2016 in Reading, UK (September 3rd and 4th). References (1) Ogawa S, Lee TM, Kay AR, Tank DW. Brain magnetic resonance imaging with contrast dependent on blood oxygenation. Proceedings of the National Academy of Sciences. 1990 Dec 1;87(24):9868-72. (2) Lindquist MA. The statistical analysis of fMRI data. Statistical Science. 2008;23 (4):439-64. doi:10.1214/09-STS282 (3) Pedoia, V., Colli, V., Strocchi, S., Vite, C., Binaghi, E. and Conte, L., 2011, March. fMRI analysis software tools: an evaluation framework. In SPIE Medical Imaging (pp. 796528-796528). International Society for Optics and Photonics. doi: 10.1117/12.877067. (4) Pedoia V, Strocchi S, Minotto R, Binaghi E. Hemispheric dominance evaluation by using fMRI activation weighted vector. Computational Modelling of Objects Represented in Images III: Fundamentals, Methods and Applications. 2012 Aug 24:303. (5) Pedoia V, Strocchi S, Colli V, Binaghi E, Conte L. Functional magnetic resonance imaging: Comparison between activation maps and computation pipelines in a clinical context. Magnetic resonance imaging. 2013 May 31;31(4):555-66. doi: 10.1016/j.mri.2012.10.013 (6) Jenkinson M, Beckmann CF, Behrens TE, Woolrich MW, Smith SM. Fsl. Neuroimage. 2012 Aug 15;62(2):782-90. doi: 10.1016/j.neuroimage.2011.09.015 (7) Duda RO, Hart PE, Stork DG. Pattern classification. John Wiley & Sons; 2012 Nov 9. (8) Mitchell TM. Machine learning. 1997. Burr Ridge, IL: McGraw Hill. 1997;45:995. (9)Finn ES, Shen X, Scheinost D, Rosenberg MD, Huang J, Chun MM, Papademetris X, Constable RT. Functional connectome fingerprinting: identifying individuals using patterns of brain connectivity. Nature neuroscience. 2015 Oct 12. doi:10.1038/nn.4135 (10) Van Den Heuvel MP, Pol HE. Exploring the brain network: a review on resting-state fMRI functional connectivity. European Neuropsychopharmacology. 2010 Aug 31;20(8):519-34. 10.1016/j.euroneuro.2010.03.008 Keywords: Brain Atlas, Functional Neuroimaging, fMRI data analysis, unsupervised learning, neuroradiology, fsl Conference: Neuroinformatics 2016, Reading, United Kingdom, 3 Sep - 4 Sep, 2016. Presentation Type: Poster Topic: Neuroimaging Citation: Vergani AA, Minotto R, Strocchi S and Binaghi E (2016). FSL-BASED HYBRID ATLAS PROMOTES ACTIVATION WEIGHTED VECTOR ANALYSIS IN FUNCTIONAL NEURORADIOLOGY. Front. Neuroinform. Conference Abstract: Neuroinformatics 2016. doi: 10.3389/conf.fninf.2016.20.00077 Copyright: The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers. They are made available through the Frontiers publishing platform as a service to conference organizers and presenters. The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated. Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed. For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions. Received: 29 May 2016; Published Online: 18 Jul 2016. * Correspondence: Dr. Alberto A Vergani, University of Insubria, Department of Theoretical and Applied Science, Varese, Italy, aavergani@uninsubria.it Login Required This action requires you to be registered with Frontiers and logged in. To register or login click here. Abstract Info Abstract The Authors in Frontiers Alberto A Vergani Renzo Minotto Sabina Strocchi Elisabetta Binaghi Google Alberto A Vergani Renzo Minotto Sabina Strocchi Elisabetta Binaghi Google Scholar Alberto A Vergani Renzo Minotto Sabina Strocchi Elisabetta Binaghi PubMed Alberto A Vergani Renzo Minotto Sabina Strocchi Elisabetta Binaghi Related Article in Frontiers Google Scholar PubMed Abstract Close Back to top Javascript is disabled. Please enable Javascript in your browser settings in order to see all the content on this page.

The present work deals with segmentation of Glial Tumors in MRI images focusing on critical aspects in manual labeling and reference estimation for segmentation validation purposes. A reproducibility analysis was conducted confirming the presence of different sources of uncertainty involved in the process of manual segmentation and responsible of high intra-operator and inter-operator variability. Technical and conceptual solutions aimed to reduce operator variability and support in the reference estimation process are integrated in GliMAn (Glial Tumor Manual Annotator), an application allowing to view and manipulate MRI volumes and implementing a label fusion strategy based on fuzzy connectedness. A set of experiments was conceived and conducted to evaluate the contribution of the solutions proposed in the process of manual segmentation and reference data estimation.

Manual MRI brain tumor segmentation is a difficult and time consuming task which makes computer support highly desirable. This paper presents a hybrid brain tumor segmentation strategy characterized by the allied use of Graph Cut segmentation method and Competitive Expectation Maximization (CEM) algorithm. Experimental results were obtained by processing in-house collected data and public data from benchmark data sets. To see if the proposed method can be considered an alternative to contemporary methods, the results obtained were compared with those obtained by authors who undertook the Multi-modal Brain Tumor Segmentation challenge. The results obtained prove that the method is competitive with recently proposed approaches.

This work focuses the attention on the automatic segmentation of meningioma from multispectral brain Magnetic Resonance imagery. The Authors address the segmentation task by proposing a fully automatic method hierarchically structured in two phases. The preliminary unsupervised phase is based on Graph Cut framework. In the second phase, preliminary segmentation results are refined using a supervised classification based on Support Vector Machine. The overall segmentation procedure is conceived fully automatic and tailored to non-volumetric data characterized by poor inter-slice spacing, in an attempt to facilitate the insertion in clinical practice. The results obtained in this preliminary study are encouraging and prove that the segmentation benefits from the allied use of Graph Cut and Support Vector Machine frameworks.