Ebook: Computational Intelligence and Bioengineering
This book is divided in the three main areas where Professor Antonina Starita was most active in the last period of her research activity: clustering and learning applications, biomedical applications, and motor control and evaluation. The part on clustering and learning applications opens with a contribution concerning the clustering of short-text corpora by Particle Swarm Optimization (PSO). The second contribution in this part investigates the use of Neural Networks (NN), and specifically of Recurrent NN, to interpret brain images obtained by functional Magnetic Resonance Imaging (fMRI). The first part of the book is closed by a contribution on the System for Paleographic Inspections (SPI) software suite.
Pisa is the town where Computer Science was born in Italy, and where many prestigious institutions in this field are located, starting from the Department of Computer Science of its University on to the Institute of Information Science and Technologies of the Italian National Research Council.
From the Seventies, there has been a hive of activity around the outstanding figure of Professor Antonina Starita (or Tonina for most of us), aimed at the development of models of information processing based on biological metaphors and at their applications in Biomedicine, involving master and doctoral students in Computer Science in Pisa, her Italian and European colleagues and their students.
These activities, at the beginning of the seventies, were still known as Cybernetics. They have then taken different names, according to the main perspective adopted, such as Bioengineering, Neural Networks, Fuzzy Logic, and Evolutionary Computation. More recently, they have one again been reunited under almost equivalent denominations, such as Soft Computing, Computational Intelligence, and Natural Computing.
Prof. Starita was born on January 31st in 1939. She studied Physics in Naples with Prof. Eduardo Caianiello, and then moved to Pisa as a researcher at the National Research ouncil, before becoming Associate Professor of Bioengineering and then Professor of Computer Science at the University of Pisa. After a long illness, Tonina left us on August 4, 2008.
Nowadays, the seminal activity of Tonina continues in the different fields of Computational Intelligence, Machine Learning and their applications to innovative interdisciplinary fields of research, through the “Computational Intelligence & Machine Learning Group” (http://www.di.unipi.it/groups/ciml) and the “Neurolab” founded by her at the Department of Computer Science of Pisa and through her alumni and collaborators outside of Pisa.
This volume, presented during a symposium in memory of Antonina Starita which was held in Pisa at the Department of Computer Science, is to be a witness as well as a big thanks to this extraordinary researcher and guide from her former students and from those who have had the privilege of meeting her in the path of their lives and to collaborate scientifically with her. It gathers scientific contributions by Tonina's alumni and main collaborators in the three main areas where Tonina was most active in the last period of her research activity: Clustering and Learning Applications, Biomedical Applications, and Motor Control and Evaluation.
The Clustering and Learning Applications part opens with a contribution by D.A. Ingaramo, M.L. Errecalde, L.C. Cagnina, and P. Rosso, concerning the clustering of short-text corpora by Particle Swarm Optimization (PSO). Short-text collections are becoming more and more frequent due to the recent development of new communication modalities, e.g. blogs, text messaging, snippets, etc., and thus it is important to develop computational tools for dealing with them. The contribution shows how PSO-based approaches can be highly competitive alternatives for clustering short-text corpora. Tonina was fascinated by PSO due to its evocation of computational mechanisms embedded in nature, and herself contributed to explore the application of this approach, e.g. to the alignment of medical images .
The second contribution in this part, by D. Sona, E. Olivetti, P. Avesani, and S. Veeramachaneni, investigates the use of Neural Networks (NN), and specifically of Recurrent NN, to interpret brain images obtained by functional Magnetic Resonance Imaging (fMRI). This technology aims to map the cognitive states of a human subject to specific functional areas of the brain. Unfortunately, the interpretation of fMRI images is not easy, and computational tools able to help in such a task are highly desirable. The authors of the contribution disclose how by using Recurrent Neural Networks they were able to win the Pittsburgh Brain Activity Interpretation Contest, edition 2006, consisting of a brain decoding problem based on free design protocol of stimuli. Tonina was very interested in Neural Networks, and in particular to the kind of models exploited by the above mentioned contribution. In fact, she contributed to the definition of a class of Neural Network models, the Recursive Neural Networks, which are generalization of Recurrent Neural Networks to the treatment of structured input, such as trees and acyclic graphs, e.g. . This class of Neural Network models is the subject of the third contribution, where A. Micheli, C. Bertinetto, C. Duce, R. Solaro, and M.R. Tiné report on the latest advances on an innovative cheminformatics approach based on them which helps into the development of new molecules and materials of biomedical interest. The contribution presents a comprehensive survey of the results obtained on acrylic and methacrylic polymers, including statistical copolymers. Previous pioneering results in the field of Cheminformatics were introduced with the contribute of Tonina, e.g. .
The first part of the book is closed by F. Aiolli and A. Ciula. In their contribution, they describe the System for Paleographic Inspections (SPI) software suite developed at the University of Pisa thanks also to Tonina's contribution . The goal of the SPI system, based on Tangent Distance and related learning methods, is to help in dating and localizing books produced by hand through the analysis of images of their ancient scripts. The authors discuss how the system has been used by paleographers in their attempts to classify and identify scripts, and how SPI can be improved further to meet the research needs of paleographers. The SPI system is another concrete example of Tonina's attitude to put innovative computational approaches at the service of other disciplines. Tonina also contributed to the development of new learning models using Tangent Distance [5,6].
The second part of this book is devoted to contributions in the medical and biological areas. These areas constituted Tonina's main research interest and much of her body of work is devoted to the definition of automatic, efficient and effective computational tools for addressing problems in these areas, with a focus on Computational Intelligent methods and techniques (e.g. [7–10]) as well as more traditional Artificial Intelligence tools, such as Expert Systems (e.g. ).
The contribution by G.C. Manikis, M.G. Kounelakis, and M. Zervakis addresses the prediction of response to induction treatment in Acute Myeloid Leukemia (AML). Different supervised learning techniques are benchmarked and evaluated and most significant indicators that contribute to the improvement of diagnosis are examined.
A further contribution to the study of the effects of treatment in AML is given by P.J.G. Lisboa, I.H. Jarman, T.A. Etchells, F. Ambrogi, I. Ardoino, M. Vignetti, and E. Biganzoli. Since AML may require aggressive systemic treatment, it is important to characterize quantitatively the response to treatment. To this aim a time-to-event model with competing risks using the framework of partial logistic artificial neural networks with automatic relevance determination (PLANNCR-ARD) is applied to a significant cohort of patients diagnosed with AML “de novo” and treated according to a strict protocol defined by the “Gruppo Italiano Malattie EMatologiche dell'Adulto” (GIMEMA) in order to follow the disease progression.
M. Pelosini, F. Baronti, and M. Petrini investigate the application of a partitioning recursive algorithm, known as Hypothesis testing Classifier System algorithm (HCS), for the characterization of patients affected by Non Hodgkin Lymphomas (NHL). The aim of the study is to discover features potentially useful to detect patients' subsets with different clinical behavior and prognosis, so to be able to personalize the treatment. In addition to that, multi-objective analysis is proposed as a tool to assess follow up schedules.
The use of Bayesian models to distinguish between benign and adnexal ovarian tumors is the topic of the contribution by B. Van Calster, O. Gevaert, C. Van Holsbeke, B. De Moor, S. Van Huffel, and D. Timmerman. The authors describe the results obtained by the many projects supported by the International Ovarian Tumor Analysis (IOTA) study group. The aim of these projects was to explore advanced mathematical modeling options for ovarian tumor diagnosis through interdisciplinary collaborations involving clinicians, statisticians, and engineers.
The study reported by P. Aretini and G. Bevilacqua focuses on the problem to automate the analysis of FISH images. A novel automated system developed by the Aristotle University of Thessaloniki with this aim is used for the evaluation of HER2 status in breast cancer cases, obtaining improved results with respect to semi-automated analysis, which has the drawback of requiring substantial user intervention.
The contribution by L. Fiaschi, J.M. Garibaldi, and N. Krasnogor investigates the possibility of discovering a correlation between variations of the individual nucleotides in DNA (Single Nucleotide Polymorphism) of a person and his/her response to drug therapy or personal susceptibility or resistance to a certain disease. Specifically, the contribution exploits the Transmission Disequilibrium Test (TDT) to perform a multiple-test analysis of SNPs for the assessment of susceptibility to pre-eclampsia. Combinations of SNPs of interest with respect to pre-eclampsia are identified.
The closing contribution of the second part reports on how new information and communications technologies, such as Computational Intelligence algorithms and tools for biodata analysis, grid computing and web services and clinical user interfaces, can be exploited to create a knowledge infrastructure to support personalised care for Alzheimer's disease (AD). The authors, E. Ifeachor, P. Hu, L. Sun, N. Hudson and M. Zervakis report the results obtained within the EU-funded project BIOPATTERN, to which Tonina actively participated. The contribution provides highlights and insights into the requirements and challenges of personalized care for AD, the characteristic features and requirements of bioprofiles within the context of AD, techniques for the acquisition of useful parameters for inclusion in the bioprofile for AD, and a grid-based prototype system to demonstrate the concepts of bioprofiling for AD within the EU setting.
The third part of the book covers a topic to which Tonina devoted a relevant portion of her research time: motor control and evaluation. Motor control was studied by Tonina in the context of Robotics (e.g. [12,13]), while motor evaluation for medical applications was the research subject of many of Tonina's projects (e.g. ).
The first contribution is by P. Morasso, V. Mohan, G. Metta, G. Sandini. They describe a method of motion planning that is based on an artificial potential field approach (Passive Motion Paradigm) combined with terminal-attractor dynamics. Besides holding interesting computational characteristics, the proposed approach addresses in a satisfactory way a feature that is crucial for complex motion patterns in humanoid robots, such as bimanual coordination or interference avoidance: precise control of the reaching time.
The second and last contribution of this part by A. Cappozzo, V. Camomilla, U. Della Croce, C. Mazzà, and G. Vannozzi, reports on the results obtained over a decade of work within the VAMA (Italian acronym for “evaluation of motor ability in the elderly”) project. The objective of VAMA was to devise, through a biomechanical analysis, quantitative methods for assessing the locomotor functional limitation of a given individual and, as a further step, to investigate the relationship between relevant impairments and disability. The contribution describes the steps constituting the successful methodology originated from this research.
The book is closed by a contribution covering an additional dimension of Tonina research activity, i.e. her effort in trying to support as much as possible the transfer of successful research products to the market, so that the quality of life of everybody could be improved in a significant way. The author of the contribution, D. Majidi, using a very personal perspective, reports on how Tonina supported Majidi's journey from her safe and comfortable lab to the difficult, and sometimes “dangerous”, world of business.
We completely agree with Majidi's statement about Tonina: “Everybody who knew her, appreciated her wisdom, her creativity, her love for life and her love for research. This article is for Tonina.”
This book is for Tonina.
Francesco Masulli, Dipartimento di Informatica e Scienze dell'Informazione, Università di Genova, Italy
Alessio Micheli, Dipartimento di Informatica, Università di Pisa, Italy
Alessandro Sperduti, Dipartimento di Matematica Pura ed Applicata, Università di Padova, Italy
The editors would like to thank the “Dipartimento di Informatica” of Pisa for their support in the organization of the symposium and the many people who have contributed to this book and to the organization of the symposium. Sincere thanks to Paulo Lisboa, Elia Biganzoli, Davide Bacciu, Umberto Barcaro, Franco Alberto Cardillo, Katuscia Cerbioni, Claudio Gallicchio, Darya Majidi, Stefania Pellegrini, and K. Brent Venable.
 G. Da San Martino, F.A. Cardillo, A. Starita. A new Swarm Intelligence Coordination Model Inspired by Collective Prey Retrieval and its Application to Image Alignment. Parallel Problem Solving from Nature (2006): 691–700.
 A. Sperduti, A. Starita. Supervised neural networks for the classification of structures. IEEE Transactions on Neural Networks 8 (1997): 714–735.
 A.M. Bianucci, A. Micheli, A. Sperduti, A. Starita, Application of cascade correlation networks for structures to chemistry, Appl. Int. 12 (2000): 117–146.
 F. Masulli, D. Sona, A. Sperduti, A. Starita, G. Zaccagnini. A System for the Automatic Morphological Analysis of Medieval Manuscripts. Journal of Forensic Document Examination 9 (1996): 45–55.
 D. Sona, A. Sperduti, and A. Starita. A Constructive Learning Algorithm for Discriminant Tangent Models. NIPS (1996): 786–792.
 D. Sona, A. Sperduti, and A. Starita. Discriminant Pattern Recognition Using Transformation Invariant Neurons. Neural Computation 12(2000):1355–1370.
 B. Rossi, F. Sartucci, A.Starita. Automatic Analysys of the Spontaneous EMG Activity During Ischaemic Test in Tetany. Electromyogr. Clin. Neurophysiol. 24 (1984): 75–80.
 S. La Manna, Darya Majidi, Antonina Starita, Davide Caramella, A. Cilotti. Magnetic resonance in mammography: a tool for the automatic detection of the regions of interest in contrast-enhanced magnetic resonance of the breast. Computer Assisted Radiology and Surgery (2001): 1275–1276.
 F.A. Cardillo, A. Starita, D. Caramella, A. Cilotti. A hybrid method for breast MR images processing and classification. IEEE International Symposium on Biomedical Imaging (2002): 181–184.
 F. Baronti, A. Micheli, A. Passaro, A. Starita. Machine Learning Contribution to Solve Prognosis Medical Problems, in Outcome Prediction in Cancer, A.F.G. Taktak and A.C. Fisher Eds., Elsevier Science, 2006.
 A. Starita, D. Majidi, A. Giordano, M. Battaglia, R. Cioni. NEUREX: A tutorial expert system for the diagnosis of neurogenic diseases of the lower limbs, Journal of Artificial Intelligence in Medicine 7 (1995): 25–36.
 F. Leoni, M. Guerrini, C. Laschi, D. Taddeucci, P. Dario, A. Starita. Implementing Robotic Grasping Tasks Using a Biological Approach. ICRA (1998): 2274–2280.
 G. Asuni, F. Leoni, E. Guglielmelli, A. Starita, P. Dario. A Neuro-controller for Robotic Manipulators Based on Biologically-Inspired Visuo-Motor Co-ordination Neural Models. International IEEE EMBS Conference on Neural Engineering (2003): 450–453.
 G. Vannozzi, U. Della Croce, A. Starita, F. Benvenuti, A. Cappozzo. Knowledge discovery in data-bases of biomechanical variables: application to the sit to stand motor task. Journal of Neuroengineering Rehabil. 1 (2004): 1–10.
Clustering of short-text collections is a very relevant research area, given the current and future mode for people to use “small-language” (e.g. blogs, snippets, news and text-message generation such as email or chat). In recent years, a few approaches based on Particle Swarm Optimization (PSO) have been proposed to solve document clustering problems. However, the particularities that arise when this kind of approaches are used for clustering corpora containing very short documents have not received too much attention by the computational linguistic community, maybe due to the high challenge that this problem implies. In this work, we propose some variants of PSO methods to deal with this kind of corpora. Our proposal includes two very different approaches to the clustering problem, which essentially differ in the representations used for maintaining the information about the clusterings under consideration. In our approach, we used two unsupervised measures of cluster validity to be optimized: the Expected Density Measure
In the last few years, functional Magnetic Resonance Imaging (fMRI) has emerged as a new and powerful method to map the cognitive states of a human subject to specific functional areas of the brain. Although fMRI has been widely used to determine average activation in different brain regions, the problem of automatically decoding the cognitive state from instantaneous brain activations has only recently been investigated. We argue that machine learning might be an effective approach to deal with fMRI image interpretation when data are collected through a free design stimulation protocol. The brain decoding task can be shaped as a classification problem. Given in input the fMRI signal of the brain, a trained model can predict the corresponding cognitive state. This study investigates the use of recurrent neural network to interpret fMRI brain images. We present the results of an empirical analysis on PBAIC data. PBAIC, namely Pittsburgh Brain Activity Interpretation Contest, proposes a brain decoding problem based on free design protocol of stimuli.
We report on the latest advances and applications of an innovative cheminformatics approach based on neural networks for structures (Recursive Neural Networks), which has recently been employed for the QSPR analysis of polymeric compounds with diverse structures. This work presents a comprehensive survey of the obtained prediction results on acrylic and methacrylic polymers, including statistical copolymers. The flexibility of the described method supports its exploitation for the study and development of new molecules and materials of biomedical interest.
Of all the disciplines that study the past through its written heritage, paleography
For an introduction to paleography see .
For an introduction to paleography see .
Acute Myeloid Leukemia is a neoplastic malignancy, which originate from the myeloid line of the cells of hematopoietic system. Although it is a relatively rare cancer and despite of improvement in prognosis due to recent radio-chemotherapy regimens, it is still a severe disease and only a minority of patients are cured. In the present work we propose a methodological approach, evaluating and benchmarking different supervised learning techniques for the prediction of response to induction treatment in Acute Myeloid Leukemia (AML). Our research also focuses on the examination of the most significant indicators that contribute to the improvement of diagnosis.
Acute Myeloid Leukaemia (AML) is a serious condition that may require aggressive systemic treatment. As a consequence of this it is important to characterize quantitatively response to treatment, differentiating patients across a range of clinical and laboratory indicators. This study follows the disease progression for a cohort of n=509 patients diagnosed with AML “de novo” and treated according to a strict protocol defined by the “Gruppo Italiano Malattie Ematologiche dell'Adulto” (GIMEMA). This protocol involves an induction therapy with health assessment typically within 60–90 days and three possible outcomes: complete remission (CR), resistance to induction therapy (Res) and induction death (ID). Accordingly, a time-to-event model with competing risks using the framework of partial logistic artificial neural networks with automatic relevance determination (PLANNCR-ARD) is applied. This results show a stratification of the mortality risk following therapy.
Non Hodgkin Lymphomas (NHL) are a group of neoplastic hematologic diseases which are characterized by chemo resistance, progressions and relapses. It would be very important to treat patients according to their disease characteristics and, when the neoplasm recurred, recognized it as soon as possible through a proper follow up program. In our study we tried to face these two aspects with the help of computer science analysis. First we developed a partitioning recursive algorithm, know as Hypothesis testing Classifier System algorithm (HCS), with the aim to discover features potentially useful to detect patients' subsets with different clinical behavior and prognosis, and therefore use it to shape the treatment. In such a way we analyzed data concerning 651 patients. The algorithm was able to detect two major groups: patients who achieved respectively complete (CR) or partial remission (PR). Even if the quality of response seemed the more important feature, when running again the algorithm others characteristics emerged among high grade NHL diagnosis, especially age and treatment approach (transplant approach and immunotherapy). Then we tried to improve our follow up schedule through a method known as multi-objective analysis. This approach works starting to choose the costs which could reflect the effectiveness of a follow up, then it calculates these values for our current one (on the basis of data available of 418 patients) and finally looks for the possible new follow up with the aim to optimized our schedule. Even if six new possible ones were obtained, the maximum improving for our current follow up was 4%: therefore our current program could be considered properly planned. In conclusion the use of this innovate approach applied to hematologic patients has been successful with the achievement of good and interesting results.
Many sophisticated methods exist to develop clinical decision support systems for daily clinical practice. In the core medical community, however, researchers often stick to basic methods due to lack of expertise. The International Ovarian Tumor Analysis (IOTA) study group, however, aims to explore advanced mathematical modeling options for ovarian tumor diagnosis through interdisciplinary collaborations involving clinicians, statisticians, and engineers. This resulted in several projects involving Bayesian models to distinguish between benign and adnexal ovarian tumors (binary classification). This chapter describes these projects. Major findings are that the classification of ovarian tumors appears to be a fairly linear problem, that benign and malignant tumors can be predicted with high accuracy, that complex black-box models can be further clarified using rule extraction, that input selection incorporating the cost of the available inputs leads to well-performing models with low total input cost, and that the widely used yet controversial and costly CA-125 tumor marker is not indispensable in mathematical diagnostic models. In conclusion, the interdisciplinary approach adopted by IOTA has resulted in useful clinical and technical insights concerning ovarian tumor diagnosis.
FISH is a direct and relatively rapid and sensitive in situ technique. No cell culture is needed in order to apply this method and results are easier to interpret than kariotype.
However, the manual evaluation of FISH image is a time consuming process prone to error involving manual counting of FISH signals over a tissue slide. Although many studies have focused on automated evaluation of FISH images, this approach remains challenging. The intensity of positive signals may be different in different experiments, even for the same sample. The differences in intensity are due to a number of factors such as the hybridization conditions and the image acquisition parameters. Many types of samples have additional complications due to the presence of cell aggregates and non uniform background fluorescence.
Therefore the FISH analysis is currently performed in a semi-automated way. The counting of dots in a semi-automated manner still remains impractical for a pathologist since it requires substantial user intervention. The Aristotle University of Thessaloniki has developed a novel automated system which aims to address these issues. The system was tested in two parallel evaluation studies at two different institutions, the University of Pisa and the Aristotle University of Thessaloniki.
The study shows that developed FISH image analysis software can impove evaluation of HER2 status in breast cancer cases.
The ability to sequence DNA has opened up the possibility to investigate if there are specific relationships between genetic information and various diseases. A variation of the individual nucleotides in DNA is called a single nucleotide polymorphism (SNP). These may have no effect or may cause subtle differences in physical or psychological characteristics. Some may actually affect a person's response to drug therapy and even confer a personal susceptibility or resistance to a certain disease. For this reason, analysis of SNPs has become the subject of extensive research. In this chapter we show an example of a multiple-test analysis of SNPs for the assessment of susceptibility to pre-eclampsia. There are many different methods that can be used for such analysis, but we focus on one known as the Transmission Disequilibrium Test (TDT). By analyzing all possible combinations of SNPs in a database related to pre-eclampsia, we identify the combinations of particular interest. This is a useful tool for the preliminary analysis of such data in order to indicate candidate SNPs for further association studies.
This paper is concerned with the exploitation of new information and communications technologies (computational intelligence algorithms and tools for biodata analysis, grid computing and web services, clinical user interfaces) to create a knowledge infrastructure to support personalised care for Alzheimer's disease (AD) - prevention, early detection, diagnosis, monitoring of progression and response to treatment. This revolves around the use of bioprofiles which is a personal ‘fingerprint’ that fuses together a person's current and past bio records and lifestyle. Analysis of an individual's bioprofile makes it possible to personalise care for AD. The paper is based on the work undertaken within the EU-funded project, BIOPATTERN. It provides highlights and insights into the requirements and challenges of personalised care for AD, the characteristic features and requirements of bioprofiles within the context of AD, techniques for the acquisition of useful parameters for inclusion in the bioprofile for AD, and a grid-based prototype system to demonstrate the concepts of bioprofiling for AD within the EU setting.
Humanoid robots have a large number of “extra” joints, organized in a humanlike fashion with several kinematic chains. In this chapter we describe a method of motion planning that is based on an artificial potential field approach (Passive Motion Paradigm) combined with terminal-attractor dynamics. No matrix inversion is necessary and the computational mechanism does not crash near kinematic singularities or when the robot is asked to achieve a final pose that is outside its intrinsic workspace: what happens, in this case, is the gentle degradation of performance that characterizes humans in the same situations. Moreover, the remaining error at equilibrium is a valuable information for triggering a reasoning process and the search of an alternative plan. The terminal attractor dynamics implicitly endows the generated trajectory with human-like smoothness and this computational framework is characterized by a feature that is crucial for complex motion patterns in humanoid robots, such as bimanual coordination or interference avoidance: precise control of the reaching time.
This paper aims at reviewing the experience maturated over the past decade by this group in the framework of the VAMA (Italian acronym for “evaluation of motor ability in the elderly”) project. The objective of that research programme was to devise quantitative methods for assessing the locomotor functional limitation of a given individual and, as a further step, for investigating the relationship between relevant impairments and disability. This was accomplished through the biomechanical analysis of a battery of selected motor tasks as executed by groups of elderly individuals. The attainment of this objective entailed the following methodological steps: increasing the resolution with which the human movement is observed; enhancing the information contained in movement-related data by applying models of the musculoskeletal system; extracting the relevant information by using specific methods of knowledge discovery from databases.
I was attending the Computer Science Department at the University of Pisa, when I met a person who was going to have a great importance in my life: Prof. Tonina Starita. I attended two of her courses at I decided that I wanted to do research in the artificial intelligence field with her. I took my degree with a thesis on an expert system for the neurological domain, and after I spent almost other four years in research in the neural networks, hybrid systems and data mining with her. She wanted me to take also my PhD, but we both understood I had entrepreneurship ambitions. We worked on a great number European Commission funded projects (among them CAMARC, WOMAN) involving different technologies (image processing, gait analysis, web ehealth services). When I announced her I was going to create a company of my own she supported and encouraged me. I am now an entrepreneur with two companies of my own. One, named Synspsis, is in the health information systems sector and the other, named DAXO, in pervasive and mobile computing, so both have been affected by my years of research with Prof. Starita, who became my friend Tonina during those years. We made research together but we had also fun together. Everybody who knows her, appreciated her wisdom, her creativity, her love for life and her love for the research. This article is for Tonina.