ABSTRACT Along with traditional morfometry methods, fractal analysis is gaining ground by the yea... more ABSTRACT Along with traditional morfometry methods, fractal analysis is gaining ground by the year among complementary tools to be used in the pathology exam of tumors.This paper reviews key progress made lately in this line of research. Difficulties that came out in the process are highlighted, as well as new insights on how to overcome them. The recent leap forward in hardware and image processing software, along with growing success of automatic and adaptive learning methods are identified as significant strengths that suggest imminent acceptance of fractal analysis in regular pathology practice.
Introduction: Virtual microscopy is expanding rapidly, based on advances in technology and availa... more Introduction: Virtual microscopy is expanding rapidly, based on advances in technology and availability. Its impact depends critically on the effective identification of regions of interest (ROI) to be thoroughly assessed. Fractal analysis is among the tools considered for preliminary evaluation of digital histological samples. Materials and Methods: Histological digital images (n 5 540) of mammary gland samples from 20 dogs and 17 cats were used to develop receiver operating characteristic (ROC) plots. The ROC models quantified the capacity of the fractal dimension of chromatin regions to classify the presence or absence of benign or malignant lesions in an image: physiologically normal tissue/adenoma/carcinoma for dogs, and physiologically normal tissue/adenoma/fibroadenoma/ carcinoma for cats. Results: There was no uniform pattern of variation of the fractal dimension of chromatin regions with malignancy, which was valid for both cats and dogs. Hence, the post-test likelihood of a lesion, as provided by the ROC model is not readily useful, especially for cats. An original fractal index for lesion highlighting was derived from the ROC model and it was proved to be a robust grading method for the probability of the presence of characteristics of malignancy in images from both dogs and cats, thus enabling the identification, in a wide histological sample, of ROI suited for further assessment. Conclusions: Using the fractal dimension of chromatin regions and statistical models, a reliable quantitative parameter was derived to enable identification of ROI in virtual microscopy histological samples.
Bulletin of University of Agricultural Sciences and Veterinary Medicine Cluj-Napoca. Veterinary Medicine, 2016
The histopathologic examination of hepatic fibrosis remains the gold standard in routine diagnosi... more The histopathologic examination of hepatic fibrosis remains the gold standard in routine diagnosis of hepatopathies in veterinary medicine. However, the visual evaluation of fibrosis is usually limited by the observer's capabilities. This paper aimed to present the applicability of fractal analysis as a quantitative method for evaluating the liver fibrosis grading using accessible tools, compared to one of the semiquantitative methods used currently in human histopathology, the METAVIR scoring system. Ovine liver histological sections taken from 22 slaughtered lambs, Masson's trichromestained, were analyzed using the METAVIR score as the semiquantitative method for staging the liver fibrosis and inflammatory reaction, and the fractal dimension of collagen regions as a quantitative method for fibrosis evaluation. The results show strong correlations between fibrosis and fractal dimension as the statistics revealed that there is a significant difference between the presence of fibrosis and its absence (p<0.001). Significant differences between the fractal dimensions corresponding to F1 and F2 fibrosis grades were observed for the images aquired with objective lens 20×. The presence of inflammatory cells did not influenced the fractal dimension of fibrosis. The Receiver Operating Characteristic (ROC) plot provided fair results for discriminating F1 versus F2 fibrosis grades, with the area under the ROC curve of 0.79. The study showed significant differences between the fractal dimension values of the liver tissue with different fibrosis grades, therefore, the fractal analysis provides a useful complementary tool in the histological examination.
Bulletin of University of Agricultural Sciences and Veterinary Medicine Cluj-Napoca. Veterinary Medicine, 2016
The histopathologic examination of hepatic fibrosis remains the gold standard in routine diagnosi... more The histopathologic examination of hepatic fibrosis remains the gold standard in routine diagnosis of hepatopathies in veterinary medicine. However, the visual evaluation of fibrosis is usually limited by the observer's capabilities. This paper aimed to present the applicability of fractal analysis as a quantitative method for evaluating the liver fibrosis grading using accessible tools, compared to one of the semiquantitative methods used currently in human histopathology, the METAVIR scoring system. Ovine liver histological sections taken from 22 slaughtered lambs, Masson's trichromestained, were analyzed using the METAVIR score as the semiquantitative method for staging the liver fibrosis and inflammatory reaction, and the fractal dimension of collagen regions as a quantitative method for fibrosis evaluation. The results show strong correlations between fibrosis and fractal dimension as the statistics revealed that there is a significant difference between the presence of fibrosis and its absence (p<0.001). Significant differences between the fractal dimensions corresponding to F1 and F2 fibrosis grades were observed for the images aquired with objective lens 20×. The presence of inflammatory cells did not influenced the fractal dimension of fibrosis. The Receiver Operating Characteristic (ROC) plot provided fair results for discriminating F1 versus F2 fibrosis grades, with the area under the ROC curve of 0.79. The study showed significant differences between the fractal dimension values of the liver tissue with different fibrosis grades, therefore, the fractal analysis provides a useful complementary tool in the histological examination.
The functional status of adypocites is reflected in their morphology and is directly related to t... more The functional status of adypocites is reflected in their morphology and is directly related to the metabolic state of the individual. Here, we investigate if fractal analysis is useful in highlighting the impact on adipocytes of progressive exposure to a hypercaloric diet. A total of 18 NMRI mice were assigned to 3 groups: control group (C), obesity induced by hypercaloric diet at one (M1) and two months (M2). Samples from mesenteric, omental, perirenal and inguinal subcutaneous adipose tissue samples were collected from each subject and analyzed through fractal dimension (FD) method. Within the performance range of current medical tests, FD showed differences between M1 and C (area below ROC curve > 0.9), as well as between M2 and C. Data collected from the inguinal subcutaneous site provided a statistical distinction between M2 and M1 (area below ROC curve 0.714). Thus, we concluded that FD represents a reliable method for identifying the smallest changes in the adipose tissue...
An artificial neural network model was developed for providing an estimation of the survival time... more An artificial neural network model was developed for providing an estimation of the survival time for cancer patients. Data from 31 dogs and cats treated as oncologic patients was used for training and validating the network which used 28 elementary predictors selected from clinical and para-clinical, macroscopic pathology, and histology information. The network was optimized to avert the overfitting, which blocks the learning process. Among the techniques tested and illustrated are: random noise in synapses, automatic cloning of well performing neurons, extended learning with/without jitter, jogging of connection weights, freezing of weights and biases, pruning of less performing nodes, cross-validation and bootstrapping methods for the selection of training and validation data sets. Once overfitting is avoided, the model provides not only reliable predictions, but also an identification of the most effective predictors.
ABSTRACT An optimization algoritm for helicopter rotors with HHC (Higher Harmonic Control), using... more ABSTRACT An optimization algoritm for helicopter rotors with HHC (Higher Harmonic Control), using aeroelastic constraints and objectives is presented inits block layout. The most time expensive part of the optimization procedure was identified to be the sensitivity analysis for aeroelastic criteria and constraints. This paper briefly describes a new package of computer programs, named HELDYN, designed for the aeroelastic analysis of non-uniform rotor blades undergoing moderate deformation. The programs are presented in their essential aspects that make them highly efficient in the sensitivity analysis. The performances of HELDYN were tested in a comprehensive parametric study that also highlighted important aspects useful for the sensitivity analysis.
A model for evaluating the survival time for dogs and cats diagnosed with mammary gland carcinoma... more A model for evaluating the survival time for dogs and cats diagnosed with mammary gland carcinoma was developed. First results are reported, based on data recorded for 74 patients, all subject to mastectomy and some of them to adjuvant chemotherapy as well. Both statistical analysis and an artificial neural network were used, integrating over 20 assumed prognostic factors, provided by clinical, pathology, and other laboratory exams. The effective operation of the statistical tools is illustrated, along with the relevant results emerging from training and validating the neural network. Fractal dimension of chromatin regions in the histology pictures from the lesion is identified among the top prognostic factors, along with tumour dimension, age, alanine transaminase and creatinine serum levels, and white blood cells count at diagnostic time. A significant impact on the survival time was also revealed for adjuvant chemotherapy, NSAID treatment, and neutering.
This study examines the hypothesis that the fractal dimension of chromatin regions in histologica... more This study examines the hypothesis that the fractal dimension of chromatin regions in histological pictures has a systematic and measurable variation when tumors emerge in a tissue. We used 194 histological pictures of healthy tissue and of tissue of the same nature with tumoral changes from 24 cases of dogs and cats treated for malignant and benign epithelial tumors: carcinoma, seminoma, adenoma, trichoblastoma, epitelioma. Fractal analysis was performed on pictures reduced to selected chromatin areas. Results indicate that for chromatin pictures at x40 magnification the fractal dimension is significantly increased when tumoral changes are present on more than 20% of the picture area. The largest effect on fractal dimension was identified for mammary gland carcinoma.
Changes in the morphology of the adypocites reflect their modified functional status, related wit... more Changes in the morphology of the adypocites reflect their modified functional status, related with the metabolic state of the individual. The current study investigates if fractal analysis is useful in highlighting the impact on adipocytes of progressive exposure to a hypercaloric diet.
An artificial neural network model was developed for providing an estimation of the survival time... more An artificial neural network model was developed for providing an estimation of the survival time for cancer patients. Data from 31 dogs and cats treated as oncologic patients was used for training and validating the network which used 28 elementary predictors selected from clinical and para-clinical, macroscopic pathology, and histology information. The network was optimized to avert the overfitting, which blocks the learning process. Among the techniques tested and illustrated are: random noise in synapses, automatic cloning of well performing neurons, extended learning with/without jitter, jogging of connection weights, freezing of weights and biases, pruning of less performing nodes, cross-validation and bootstrapping methods for the selection of training and validation data sets. Once overfitting is avoided, the model provides not only reliable predictions, but also an identification of the most effective predictors.
An optimization algoritm for helicopter rotors with HHC (Higher Harmonic Control), using aeroelas... more An optimization algoritm for helicopter rotors with HHC (Higher Harmonic Control), using aeroelastic constraints and objectives is presented inits block layout. The most time expensive part of the optimization procedure was identified to be the sensitivity analysis for aeroelastic criteria and constraints. This paper briefly describes a new package of computer programs, named HELDYN, designed for the aeroelastic analysis of non-uniform rotor blades undergoing moderate deformation. The programs are presented in their essential aspects that make them highly efficient in the sensitivity analysis. The performances of HELDYN were tested in a comprehensive parametric study that also highlighted important aspects useful for the sensitivity analysis.
This study examines the hypothesis that the fractal dimension of chromatin regions in histologica... more This study examines the hypothesis that the fractal dimension of chromatin regions in histological pictures has a systematic and measurable variation when tumors emerge in a tissue. We used 194 histological pictures of healthy tissue and of tissue of the same nature with tumoral changes from 24 cases of dogs and cats treated for malignant and benign epithelial tumors: carcinoma, seminoma, adenoma, trichoblastoma, epitelioma. Fractal analysis was performed on pictures reduced to selected chromatin areas. Results indicate that for chromatin pictures at x40 magnification the fractal dimension is significantly increased when tumoral changes are present on more than 20% of the picture area. The largest effect on fractal dimension was identified for mammary gland carcinoma.
Stadiul actual al aplicării analizei fractale în diagnosticul şi prognosticul tumorilor 2/25 Stad... more Stadiul actual al aplicării analizei fractale în diagnosticul şi prognosticul tumorilor 2/25 Stadiul actual al aplicării analizei fractale în diagnosticul şi prognosticul tumorilor 3/25 Aşa sunt, de fapt, toate structurile biologice, care pot fi uşor identificate ca fractali aproximativi [Figura 2]. 2. Ce este analiza fractală Analiza fractală este ansamblul metodelor de organizare, prelucrare şi interpretare a informaŃiei disponibile despre caracteristicile fractale ale unui obiect: complexitatea şi auto-invarianŃa. Stadiul actual al aplicării analizei fractale în diagnosticul şi prognosticul tumorilor 24/25
ABSTRACT Along with traditional morfometry methods, fractal analysis is gaining ground by the yea... more ABSTRACT Along with traditional morfometry methods, fractal analysis is gaining ground by the year among complementary tools to be used in the pathology exam of tumors.This paper reviews key progress made lately in this line of research. Difficulties that came out in the process are highlighted, as well as new insights on how to overcome them. The recent leap forward in hardware and image processing software, along with growing success of automatic and adaptive learning methods are identified as significant strengths that suggest imminent acceptance of fractal analysis in regular pathology practice.
Introduction: Virtual microscopy is expanding rapidly, based on advances in technology and availa... more Introduction: Virtual microscopy is expanding rapidly, based on advances in technology and availability. Its impact depends critically on the effective identification of regions of interest (ROI) to be thoroughly assessed. Fractal analysis is among the tools considered for preliminary evaluation of digital histological samples. Materials and Methods: Histological digital images (n 5 540) of mammary gland samples from 20 dogs and 17 cats were used to develop receiver operating characteristic (ROC) plots. The ROC models quantified the capacity of the fractal dimension of chromatin regions to classify the presence or absence of benign or malignant lesions in an image: physiologically normal tissue/adenoma/carcinoma for dogs, and physiologically normal tissue/adenoma/fibroadenoma/ carcinoma for cats. Results: There was no uniform pattern of variation of the fractal dimension of chromatin regions with malignancy, which was valid for both cats and dogs. Hence, the post-test likelihood of a lesion, as provided by the ROC model is not readily useful, especially for cats. An original fractal index for lesion highlighting was derived from the ROC model and it was proved to be a robust grading method for the probability of the presence of characteristics of malignancy in images from both dogs and cats, thus enabling the identification, in a wide histological sample, of ROI suited for further assessment. Conclusions: Using the fractal dimension of chromatin regions and statistical models, a reliable quantitative parameter was derived to enable identification of ROI in virtual microscopy histological samples.
Bulletin of University of Agricultural Sciences and Veterinary Medicine Cluj-Napoca. Veterinary Medicine, 2016
The histopathologic examination of hepatic fibrosis remains the gold standard in routine diagnosi... more The histopathologic examination of hepatic fibrosis remains the gold standard in routine diagnosis of hepatopathies in veterinary medicine. However, the visual evaluation of fibrosis is usually limited by the observer's capabilities. This paper aimed to present the applicability of fractal analysis as a quantitative method for evaluating the liver fibrosis grading using accessible tools, compared to one of the semiquantitative methods used currently in human histopathology, the METAVIR scoring system. Ovine liver histological sections taken from 22 slaughtered lambs, Masson's trichromestained, were analyzed using the METAVIR score as the semiquantitative method for staging the liver fibrosis and inflammatory reaction, and the fractal dimension of collagen regions as a quantitative method for fibrosis evaluation. The results show strong correlations between fibrosis and fractal dimension as the statistics revealed that there is a significant difference between the presence of fibrosis and its absence (p<0.001). Significant differences between the fractal dimensions corresponding to F1 and F2 fibrosis grades were observed for the images aquired with objective lens 20×. The presence of inflammatory cells did not influenced the fractal dimension of fibrosis. The Receiver Operating Characteristic (ROC) plot provided fair results for discriminating F1 versus F2 fibrosis grades, with the area under the ROC curve of 0.79. The study showed significant differences between the fractal dimension values of the liver tissue with different fibrosis grades, therefore, the fractal analysis provides a useful complementary tool in the histological examination.
Bulletin of University of Agricultural Sciences and Veterinary Medicine Cluj-Napoca. Veterinary Medicine, 2016
The histopathologic examination of hepatic fibrosis remains the gold standard in routine diagnosi... more The histopathologic examination of hepatic fibrosis remains the gold standard in routine diagnosis of hepatopathies in veterinary medicine. However, the visual evaluation of fibrosis is usually limited by the observer's capabilities. This paper aimed to present the applicability of fractal analysis as a quantitative method for evaluating the liver fibrosis grading using accessible tools, compared to one of the semiquantitative methods used currently in human histopathology, the METAVIR scoring system. Ovine liver histological sections taken from 22 slaughtered lambs, Masson's trichromestained, were analyzed using the METAVIR score as the semiquantitative method for staging the liver fibrosis and inflammatory reaction, and the fractal dimension of collagen regions as a quantitative method for fibrosis evaluation. The results show strong correlations between fibrosis and fractal dimension as the statistics revealed that there is a significant difference between the presence of fibrosis and its absence (p<0.001). Significant differences between the fractal dimensions corresponding to F1 and F2 fibrosis grades were observed for the images aquired with objective lens 20×. The presence of inflammatory cells did not influenced the fractal dimension of fibrosis. The Receiver Operating Characteristic (ROC) plot provided fair results for discriminating F1 versus F2 fibrosis grades, with the area under the ROC curve of 0.79. The study showed significant differences between the fractal dimension values of the liver tissue with different fibrosis grades, therefore, the fractal analysis provides a useful complementary tool in the histological examination.
The functional status of adypocites is reflected in their morphology and is directly related to t... more The functional status of adypocites is reflected in their morphology and is directly related to the metabolic state of the individual. Here, we investigate if fractal analysis is useful in highlighting the impact on adipocytes of progressive exposure to a hypercaloric diet. A total of 18 NMRI mice were assigned to 3 groups: control group (C), obesity induced by hypercaloric diet at one (M1) and two months (M2). Samples from mesenteric, omental, perirenal and inguinal subcutaneous adipose tissue samples were collected from each subject and analyzed through fractal dimension (FD) method. Within the performance range of current medical tests, FD showed differences between M1 and C (area below ROC curve > 0.9), as well as between M2 and C. Data collected from the inguinal subcutaneous site provided a statistical distinction between M2 and M1 (area below ROC curve 0.714). Thus, we concluded that FD represents a reliable method for identifying the smallest changes in the adipose tissue...
An artificial neural network model was developed for providing an estimation of the survival time... more An artificial neural network model was developed for providing an estimation of the survival time for cancer patients. Data from 31 dogs and cats treated as oncologic patients was used for training and validating the network which used 28 elementary predictors selected from clinical and para-clinical, macroscopic pathology, and histology information. The network was optimized to avert the overfitting, which blocks the learning process. Among the techniques tested and illustrated are: random noise in synapses, automatic cloning of well performing neurons, extended learning with/without jitter, jogging of connection weights, freezing of weights and biases, pruning of less performing nodes, cross-validation and bootstrapping methods for the selection of training and validation data sets. Once overfitting is avoided, the model provides not only reliable predictions, but also an identification of the most effective predictors.
ABSTRACT An optimization algoritm for helicopter rotors with HHC (Higher Harmonic Control), using... more ABSTRACT An optimization algoritm for helicopter rotors with HHC (Higher Harmonic Control), using aeroelastic constraints and objectives is presented inits block layout. The most time expensive part of the optimization procedure was identified to be the sensitivity analysis for aeroelastic criteria and constraints. This paper briefly describes a new package of computer programs, named HELDYN, designed for the aeroelastic analysis of non-uniform rotor blades undergoing moderate deformation. The programs are presented in their essential aspects that make them highly efficient in the sensitivity analysis. The performances of HELDYN were tested in a comprehensive parametric study that also highlighted important aspects useful for the sensitivity analysis.
A model for evaluating the survival time for dogs and cats diagnosed with mammary gland carcinoma... more A model for evaluating the survival time for dogs and cats diagnosed with mammary gland carcinoma was developed. First results are reported, based on data recorded for 74 patients, all subject to mastectomy and some of them to adjuvant chemotherapy as well. Both statistical analysis and an artificial neural network were used, integrating over 20 assumed prognostic factors, provided by clinical, pathology, and other laboratory exams. The effective operation of the statistical tools is illustrated, along with the relevant results emerging from training and validating the neural network. Fractal dimension of chromatin regions in the histology pictures from the lesion is identified among the top prognostic factors, along with tumour dimension, age, alanine transaminase and creatinine serum levels, and white blood cells count at diagnostic time. A significant impact on the survival time was also revealed for adjuvant chemotherapy, NSAID treatment, and neutering.
This study examines the hypothesis that the fractal dimension of chromatin regions in histologica... more This study examines the hypothesis that the fractal dimension of chromatin regions in histological pictures has a systematic and measurable variation when tumors emerge in a tissue. We used 194 histological pictures of healthy tissue and of tissue of the same nature with tumoral changes from 24 cases of dogs and cats treated for malignant and benign epithelial tumors: carcinoma, seminoma, adenoma, trichoblastoma, epitelioma. Fractal analysis was performed on pictures reduced to selected chromatin areas. Results indicate that for chromatin pictures at x40 magnification the fractal dimension is significantly increased when tumoral changes are present on more than 20% of the picture area. The largest effect on fractal dimension was identified for mammary gland carcinoma.
Changes in the morphology of the adypocites reflect their modified functional status, related wit... more Changes in the morphology of the adypocites reflect their modified functional status, related with the metabolic state of the individual. The current study investigates if fractal analysis is useful in highlighting the impact on adipocytes of progressive exposure to a hypercaloric diet.
An artificial neural network model was developed for providing an estimation of the survival time... more An artificial neural network model was developed for providing an estimation of the survival time for cancer patients. Data from 31 dogs and cats treated as oncologic patients was used for training and validating the network which used 28 elementary predictors selected from clinical and para-clinical, macroscopic pathology, and histology information. The network was optimized to avert the overfitting, which blocks the learning process. Among the techniques tested and illustrated are: random noise in synapses, automatic cloning of well performing neurons, extended learning with/without jitter, jogging of connection weights, freezing of weights and biases, pruning of less performing nodes, cross-validation and bootstrapping methods for the selection of training and validation data sets. Once overfitting is avoided, the model provides not only reliable predictions, but also an identification of the most effective predictors.
An optimization algoritm for helicopter rotors with HHC (Higher Harmonic Control), using aeroelas... more An optimization algoritm for helicopter rotors with HHC (Higher Harmonic Control), using aeroelastic constraints and objectives is presented inits block layout. The most time expensive part of the optimization procedure was identified to be the sensitivity analysis for aeroelastic criteria and constraints. This paper briefly describes a new package of computer programs, named HELDYN, designed for the aeroelastic analysis of non-uniform rotor blades undergoing moderate deformation. The programs are presented in their essential aspects that make them highly efficient in the sensitivity analysis. The performances of HELDYN were tested in a comprehensive parametric study that also highlighted important aspects useful for the sensitivity analysis.
This study examines the hypothesis that the fractal dimension of chromatin regions in histologica... more This study examines the hypothesis that the fractal dimension of chromatin regions in histological pictures has a systematic and measurable variation when tumors emerge in a tissue. We used 194 histological pictures of healthy tissue and of tissue of the same nature with tumoral changes from 24 cases of dogs and cats treated for malignant and benign epithelial tumors: carcinoma, seminoma, adenoma, trichoblastoma, epitelioma. Fractal analysis was performed on pictures reduced to selected chromatin areas. Results indicate that for chromatin pictures at x40 magnification the fractal dimension is significantly increased when tumoral changes are present on more than 20% of the picture area. The largest effect on fractal dimension was identified for mammary gland carcinoma.
Stadiul actual al aplicării analizei fractale în diagnosticul şi prognosticul tumorilor 2/25 Stad... more Stadiul actual al aplicării analizei fractale în diagnosticul şi prognosticul tumorilor 2/25 Stadiul actual al aplicării analizei fractale în diagnosticul şi prognosticul tumorilor 3/25 Aşa sunt, de fapt, toate structurile biologice, care pot fi uşor identificate ca fractali aproximativi [Figura 2]. 2. Ce este analiza fractală Analiza fractală este ansamblul metodelor de organizare, prelucrare şi interpretare a informaŃiei disponibile despre caracteristicile fractale ale unui obiect: complexitatea şi auto-invarianŃa. Stadiul actual al aplicării analizei fractale în diagnosticul şi prognosticul tumorilor 24/25
The impact that progress in information technology has on the entire activity of humans has propo... more The impact that progress in information technology has on the entire activity of humans has proportions and a complexity that are difficult to describe or summarize. On one hand, the objects built by humans become – more and more each day – “smart”, or at least capable of collecting, processing and transmitting information. However, these objects are increasingly capable of making decisions, based on pre-programmed algorithms or based on heuristic “learning” techniques – through progressive exposure to the environment and the preferences and options of the user. On the other hand, all of the intellectual and even emotional activity of modern man is increasingly assisted and dependent on instruments developed through information technology: databases, search engines, communication devices, operational models, heuristic models.
This progress in information technology has submitted the actual paradigm of scientific knowledge to revision pressures unprecedented from Karl Popper to the present day. The new debate has coagulated like a show around the confrontation between Noam Chomski, the illustrious American linguist and political scientist and Peter Norvig, the leader of the research department in Google company. This confrontation of ideas began in 2011, at the Brains, Minds, and Machines symposium hosted by the Massachusetts Institute of Technology, from „how one can obtain artificial intelligence” and, in a few months, through articles and relatively elaborate interventions [Norvig, 2012] became about „what human knowledge is”. Essentially, it is about knowledge understood as an explicative model, simple enough for every link to be logically comprehensible (Chomski) and, alternatively, knowledge as an informational model, a „black box”, capable of offering increasingly better predictions based on historical data and internal statistical search algorithms (Norvig). The question is: does the progress of knowledge still depend on models that are elegant in their simplicity, logically accessible – in integrity and in detail, models that are continually reviewed to approximate the available data as well as possible? Or are we entering a new era of statistical informational models, capable of continuously processing enormous quantities of data – historical or real time data and providing classifications, predictions, etc. – but without subjacent explanations: we find out how much, when, with associated probabilities, but the question „why?” loses its relevance.
In the applied medical sciences, the most significant reflection of this pressure to change the relationship paradigm between historical data – formalized knowledge – the practice of decisions and predictions is Evidence Based Medicine, EBM. It is not a new concept – its origins can be traced (at least) to the comparative clinical study on six groups of patients, through which the British physician James Lind, of the Royal Navy, established in 1747 what the foods than can prevent and treat scurvy – lemons and oranges – were. Isolating vitamin C and clarifying the pathogenesis of scurvy came about almost 200 years later, in 1932, but the prophylactic treatment had already been generalized by the Royal Navy as early as 1787, based on the evidence collected by Lind. The reference definition of EBM was settled in 1996: „the conscious, explicit and judicious use of the best evidence in order to formulate decisions with regard to the treatment of individual patients” [Sackett, 1996]. Best evidence meaning proof (as certain as possible), not explanations (as plausible as possible). In the last few decades, the scope of medical research and the increasing pressure for the maximum but responsible use of the results of this research imposed EBM as a dominant technique for collecting, organizing, evaluating and using medical information; EBM procedures are now standardized at a national and international level, with specialized associations and regulation and certification organizations facilitating the development of work methods [Mayer, 2010].
In a good correlation with EBM, a direction of accelerated progress in applying new information techniques in medicine is that of prediction models capable of continuous development, through what is suggestively known as „learning” or „training”. Artificial neural networks are one of the forms that were intensely studied and tested in this context, as part of the research surrounding „artificial intelligence”. Promising results have been reported more and more frequently in applying neural networks to diagnosis [Ganesan, 2010], [Jiang, 2010], [Barwad, 2012], [Amato, 2013], but also to evaluating prognosis and prediction factors [Lundin, 1999]. The number of such factors under evaluation has risen exponentially lately, especially due to the progress in genomics and molecular biology, which gave rise to a necessity for new information instruments, capable of processing such a large data volume. Prediction based on such information models supplied with huge databases is already transitioning into the current medical practice [Silva, 2015], [Prigg, 2015]: at Beth Israel Deaconess Hospital in Boston USA, Dr. Steven Horng recently announced the operational phase for of an information instrument that allows the formulation of a diagnosis, indicates recommended therapeutic procedures and estimates the prognosis for every patient in intensive therapy. Dr. Horng, a physician specializing in both emergency medicine and biomedical IT, believes that the accuracy with which the program formulates, for example, the prediction for the imminence of the patient’s death in the next 30 days is over 96%. The recent and historical clinical and paraclinical data of the patient are introduced into the information model based on similar historical data collected in the last 30 years from over 250000 patients. The database is augmented continuously through the regular collection of information, every 3 minutes, from the physiological data monitors installed for all the patients admitted into the ICU of the hospital.
In the last 30 years, technological progress has brought about a true revolution in medical imagery [Dougherty, 2009]. On one hand, minimally invasive investigation techniques have become accessible for a very large number of patients, both in human and veterinary medicine: MRIs, computerized tomography, nuclear magnetic resonance, confocal reflection microscopy. On the other hand, the increased performance of information instruments facilitates new approaches in distinguishing useful information in medical images [Bubnov, 2014], [Lennon, 2015]. The digital format, practically generalized both for the results in new investigation techniques, as well as for classical microscopy and radiology, now facilitates processing for the extraction of useful information, storing and accessing images, as well as the comparison, morphometric and statistical analysis, identification of regions of interest/lesions etc. [Rangayyan, 2005].
The road towards semi-automated and even automated procedures that can be used directly in research, screening, pre-diagnosis or diagnosis is open and is beginning to be intensely exploited, including by the commercial component of medical activity. Using automated and semi-automated procedures also leads to an exponential increase in the volume of collected data which, in turn, reclaims information approaches such as neural networks for processes or such as EBM for testing and application. It is a progress spiral that is presently accelerating and in which man, the professional, the expert, can be centrifugally marginalized towards positions in which he only contributes punctually to ample, interdependent processes that he does not control anymore. It is possible that we are witnessing the dawn of a new historical era, in which the virtual reality created continuously by humans becomes as significant and autonomous as objective reality. Our attempt to know and change reality in its entirety for our own purposes shall inevitably adapt to the new situation.
Simon Rosenfeld, from the National Cancer Institute – USA, considers that oncology and oncological research are at a critical crossroads with the dynamics of nonlinear systems and research regarding swarm intelligence in the contemporary attempt to offer a new, unifying vision with regard to the way in which systems as diverse as „biological networks, communities of social insects, robotic communities, microbial communities, communities of somatic cells, (…) social networks, and (…) many other systems” [Rosenfeld, 2015] appear and function. Rosenfeld’s unifying vision marks an essential link between carcinogenesis, fractal analysis – that was mainly developed as a modeling instrument for nonlinear dynamic systems and artificial neural networks – that are one of the best documented examples of man-made swarm intelligence.
Fractal analysis facilitates the obtaining of measurements, of numbers that quantify the non-periodical complexity that is so characteristic of biological structures, on all size scales in which life is manifested. This research has investigated the possibility of including fractal analysis applied to histopathological or cytopathological microscopic images amongst the instruments used for diagnosing tumours in animals. I then tested the introduction of the fractal dimension amongst oncological prognostic factors. For this purpose, I developed statistical models and artificial neural network-type models for the survival duration, including the fractal dimension in the set of predictors based on which these models estimate the prognostic for cancer patients.
Materials and methods For 540 histological digital images of mammary gland samples (×40, HE stain... more Materials and methods For 540 histological digital images of mammary gland samples (×40, HE stain), from 20 dogs and 17 cats, the fractal dimension of chromatin regions (FDCR) was computed and used to develop Receiver Operating Characteristic (ROC) plots. The ROC models quantified the capacity of FDCR to classify the presence or absence of benign or malign lesions in a picture: physiologically normal tissue / adenoma / carcinoma – for dogs, and physiologically normal tissue / adenoma / fibroadenoma / carcinoma for cats. All the images selected for the study were of normal mammary tissue or mammary tissue that was tumorally modified exclusively at the level of the epithelial components. The study did not include images that had a mixed content of pathological characters, with the presence in significant weights of several types of lesions. The pool of images was used as a model for local frame pictures from virtual microscopy scans, to test procedures for identifying Regions of Interest ROI, those frames where the probability for the presence of malignant lesions is higher. Image processing included complex balance and sharpening. The chromatin regions were extracted by colour maps. The fractal dimension of grey images was calculated by the box method. The main software packages that were used are: Olympus Cell^B for capturing images, CorelPHOTO-PAINT ® for the preliminary processing of images, FracLab 2.05 for fractal analysis, StatsDirect 3.0 for the statistical analysis and graphical representation of the results, Epi Info TM 7 for the database of cases. Results While in dogs, FDCR increases with the malignancy of lesions, in cats the relation is less regular. To overcome this problem, a newly created Fractal Index for Lesion Highlighting – FILH, derived from the ROC statistic model was tested and demonstrated to provide a robust orientation with regard to the presence/absence of malignity features in a histological image, for both dogs and cats. FILH = (post_test likelihood of lesion for positive test + post_test likelihood of lesion despite negative test) / 2 ESVP-ECVP Annual Meeting Bologna 2016 References: 1. Gaiță L., Applications Of Fractal Analysis In The Anatomopathological Examination Of Tumor Processes, PhD Thesis, Scientific coordinator Prof. dr. M. Militaru, University of Agronomic Sciences and Veterinary Medicine of Bucharest, 2016 2. Gaiță L, Militaru M, 2010, Current situation in using fractal analysis for tumor diagnostic and prognostic, Revista Romana de Medicina Veterinară, ISSN 1220-3173; 20(3) pg. 11-30 2. Zweig M. H., Campbell G., 1993, Receiver-operating characteristic (ROC) plots: a fundamental evaluation tool in clinical medicine, Clinical Chemistry, Vol. 39:4, pg. 561-77 7
Uploads
Papers by Liviu Gaita
This progress in information technology has submitted the actual paradigm of scientific knowledge to revision pressures unprecedented from Karl Popper to the present day. The new debate has coagulated like a show around the confrontation between Noam Chomski, the illustrious American linguist and political scientist and Peter Norvig, the leader of the research department in Google company. This confrontation of ideas began in 2011, at the Brains, Minds, and Machines symposium hosted by the Massachusetts Institute of Technology, from „how one can obtain artificial intelligence” and, in a few months, through articles and relatively elaborate interventions [Norvig, 2012] became about „what human knowledge is”. Essentially, it is about knowledge understood as an explicative model, simple enough for every link to be logically comprehensible (Chomski) and, alternatively, knowledge as an informational model, a „black box”, capable of offering increasingly better predictions based on historical data and internal statistical search algorithms (Norvig). The question is: does the progress of knowledge still depend on models that are elegant in their simplicity, logically accessible – in integrity and in detail, models that are continually reviewed to approximate the available data as well as possible? Or are we entering a new era of statistical informational models, capable of continuously processing enormous quantities of data – historical or real time data and providing classifications, predictions, etc. – but without subjacent explanations: we find out how much, when, with associated probabilities, but the question „why?” loses its relevance.
In the applied medical sciences, the most significant reflection of this pressure to change the relationship paradigm between historical data – formalized knowledge – the practice of decisions and predictions is Evidence Based Medicine, EBM. It is not a new concept – its origins can be traced (at least) to the comparative clinical study on six groups of patients, through which the British physician James Lind, of the Royal Navy, established in 1747 what the foods than can prevent and treat scurvy – lemons and oranges – were. Isolating vitamin C and clarifying the pathogenesis of scurvy came about almost 200 years later, in 1932, but the prophylactic treatment had already been generalized by the Royal Navy as early as 1787, based on the evidence collected by Lind. The reference definition of EBM was settled in 1996:
„the conscious, explicit and judicious use of the best evidence in order to formulate decisions with regard to the treatment of individual patients” [Sackett, 1996]. Best evidence meaning proof (as certain as possible), not explanations (as plausible as possible). In the last few decades, the scope of medical research and the increasing pressure for the maximum but responsible use of the results of this research imposed EBM as a dominant technique for collecting, organizing, evaluating and using medical information; EBM procedures are now standardized at a national and international level, with specialized associations and regulation and certification organizations facilitating the development of work methods [Mayer, 2010].
In a good correlation with EBM, a direction of accelerated progress in applying new information techniques in medicine is that of prediction models capable of continuous development, through what is suggestively known as „learning” or „training”. Artificial neural networks are one of the forms that were intensely studied and tested in this context, as part of the research surrounding „artificial intelligence”. Promising results have been reported more and more frequently in applying neural networks to diagnosis [Ganesan, 2010], [Jiang, 2010], [Barwad, 2012], [Amato, 2013], but also to evaluating prognosis and prediction factors [Lundin, 1999]. The number of such factors under evaluation has risen exponentially lately, especially due to the progress in genomics and molecular biology, which gave rise to a necessity for new information instruments, capable of processing such a large data volume. Prediction based on such information models supplied with huge databases is already transitioning into the current medical practice [Silva, 2015], [Prigg, 2015]: at Beth Israel Deaconess Hospital in Boston USA, Dr. Steven Horng recently announced the operational phase for of an information instrument that allows the formulation of a diagnosis, indicates recommended therapeutic procedures and estimates the prognosis for every patient in intensive therapy. Dr. Horng, a physician specializing in both emergency medicine and biomedical IT, believes that the accuracy with which the program formulates, for example, the prediction for the imminence of the patient’s death in the next 30 days is over 96%. The recent and historical clinical and paraclinical data of the patient are introduced into the information model based on similar historical data collected in the last 30 years from over 250000 patients. The database is augmented continuously through the regular collection of information, every 3 minutes, from the physiological data monitors installed for all the patients admitted into the ICU of the hospital.
In the last 30 years, technological progress has brought about a true revolution in medical imagery [Dougherty, 2009]. On one hand, minimally invasive investigation techniques have become accessible for a very large number of patients, both in human and veterinary medicine: MRIs, computerized tomography, nuclear magnetic resonance, confocal reflection microscopy. On the other hand, the increased performance of information instruments facilitates new approaches in distinguishing useful information in medical images [Bubnov, 2014], [Lennon, 2015]. The digital format, practically generalized both for the results in new investigation techniques, as well as for classical microscopy and radiology, now facilitates processing for the extraction of useful information, storing and accessing images, as well as the comparison, morphometric and statistical analysis, identification of regions of interest/lesions etc. [Rangayyan, 2005].
The road towards semi-automated and even automated procedures that can be used directly in research, screening, pre-diagnosis or diagnosis is open and is beginning to be intensely exploited, including by the commercial component of medical activity. Using automated and semi-automated procedures also leads to an exponential increase in the volume of collected data which, in turn, reclaims information approaches such as neural networks for processes or such as EBM for testing and application. It is a progress spiral that is presently accelerating and in which man, the professional, the expert, can be centrifugally marginalized towards positions in which he only contributes punctually to ample, interdependent processes that he does not control anymore. It is possible that we are witnessing the dawn of a new historical era, in which the virtual reality created continuously by humans becomes as significant and autonomous as objective reality. Our attempt to know and change reality in its entirety for our own purposes shall inevitably adapt to the new situation.
Simon Rosenfeld, from the National Cancer Institute – USA, considers that oncology and oncological research are at a critical crossroads with the dynamics of nonlinear systems and research regarding swarm intelligence in the contemporary attempt to offer a new, unifying vision with regard to the way in which systems as diverse as „biological networks, communities of social insects, robotic communities, microbial communities, communities of somatic cells, (…) social networks, and (…) many other systems” [Rosenfeld, 2015] appear and function. Rosenfeld’s unifying vision marks an essential link between carcinogenesis, fractal analysis – that was mainly developed as a modeling instrument for nonlinear dynamic systems and artificial neural networks – that are one of the best documented examples of man-made swarm intelligence.
Fractal analysis facilitates the obtaining of measurements, of numbers that quantify the non-periodical complexity that is so characteristic of biological structures, on all size scales in which life is manifested. This research has investigated the possibility of including fractal analysis applied to histopathological or cytopathological microscopic images amongst the instruments used for diagnosing tumours in animals. I then tested the introduction of the fractal dimension amongst oncological prognostic factors. For this purpose, I developed statistical models and artificial neural network-type models for the survival duration, including the fractal dimension in the set of predictors based on which these models estimate the prognostic for cancer patients.