V. Zeljkovic

Automated adipose tissue detection method

Obesity or adiposity is a major global health problem. Adipose tissue has been subdivided into two types: white adipose tissue (WAT) that stores energy, and brown adipose tissue (BAT) that functions to dissipate energy in the form of heat. BAT plays a particularly important role in newborns, which use this tissue to defend themselves against cold. However, using glucose tracer, it has been independently demonstrated that metabolically active BAT exists in adults. This has revived a scientific interest in BAT, targeting BAT as a new method to reduce obesity. However, BAT is extremely difficult to visualize, and even more to quantify. That is why we developed a mathematical model and intelligent software capable of automatically detecting and analyzing BAT, with the goal of quantifying it in the mammary adipose microenvironment. We utilized an animal model with abnormal mammary adipose tissue environment with increased amount of BAT in comparison with normal animals for this analysis. The proposed algorithm has two-fold function: visual and numerical. Visual function is expressed through detection and indication of the potential brown fat on the tissue microscopic images by marking it in a different color and distinguishing it from the rest of the tissue. Numerical function enables quantification of the amount of detected brown fat by calculating its numerical equivalent.

Improved algorithm for mammary adipose microenvironment definition byl automated brown fat quantification

Adipose tissue, known also as fat, is a loose connective tissue that fills up space between the organs and provides structural and metabolic support. The classical functions of adipose tissue are the storage of energy in the form of triglycerides, and thermal insulation. Historically, adipose tissue has been subdivided into two main types: white fat that stores energy, and brown fat, that induce thermogenesis and produce heat. Brown fat was previously regarded as a special type of fat relevant only for newborns and hibernating animals defending against cold. Recently has been shown that brown fat is physiologically present and active in adult humans. However, brown fat is extremely difficult to visualize and even more difficult to quantify. We have found that levels of brown fat are dramatically increased in mammary tissue of the BRCA1 mutant mice compared to the normal, wild-type mice. We attempted to quantify the amount of brown fat in histological sections of mammary tissues from these mice. We initially presented a mathematical model and intelligent software capable of automatically detecting and quantifying the content of fat adipose tissue, with the goal of defining the mammary adipose microenvironment. The proposed algorithm showed itself to be very successful in detecting brown adipose tissue in images that contain pure mammary tissue. We propose here improved adipose tissue detection algorithm capable of successfully detecting brown fat in the whole mammary gland image that contains extra information like arteries, background and other surrounding elements. The proposed method has two fold function: visual and numerical. Visual function is expressed through detection and indication of the potential brown fat in the tissue microscopic images by marking it in a different color and distinguishing it from the rest of the tissue. Numerical function enables quantification of the amount of detected brown fat by calculating its numerical content equivalent.

Supplemental melanoma diagnosis for darker skin complexion gradients

Melanoma represents one of most malignant tumors associated with melanocytes in pigmented cells of the skin and in particular is a result of malignant transformation of melanocytes. Due to migration of neural cell crest, melanoma can develop not only on skin, but on oral and genital mucosa, and also gastrointestinal tract and brain. Melanoma is usually present and manifests itself with changes in color, size, contour and configuration, or may occur as new pigmented lesions. In particular, melanoma represents the sixth leading cause of malignancy in the United States with much higher mortality rate among non-Caucasian population, although is more common among whites. Considering its complexity, clinical diagnosis of melanoma is challenging even for experienced dermatologists. This is why it is necessary to develop computer assisted diagnostic tool for melanoma detection focused on dark and fair complexion skin which adds more objective judgments based on quantitative measures. Therefore, specialized algorithms were developed and tested utilizing databases including images of a variety of skin cancer manifestations. Those diagnostic indicators were assessed utilizing commonly used ABCDE criteria for different skin complexions and also natural and simulated darker background reflecting darker skin tones associated with different ethnic groups. Incorporated Canny, Prewitt, Roberts and Sobel edge detectors allowed to optimize melanoma diagnosis for darker skin tones and assess the degree of correct classification for each of ABCDE criterion reflecting varied skin complexion.

Mathematical models for bone density assessment

A decrease in bone strength associated with osteoporosis and resulting increased susceptibility to fracture continue to be one of the critical challenges of aging population. It is estimated that over 200 million people worldwide suffer from these conditions. Due to difficulty in accurate evaluation of a bone loss, novel approaches in assessing bone overall integrity are introduced. These approaches included the use of micro CT scan bone images of mouse which allow to study simulated related bone structure changes and their modelling applying GMM models.

Quantitative assessment of progression/regression stages in severe dermatologic manifestations

Varicella-zoster virus causes chickenpox, a very contagious disease manifested by rash, itching, tiredness, and fever. The affected individual gets stomach, back and face covered with blister-like rash that often spreads over the entire body resulting in 250 or even 500 itchy blisters. Babies, adults, and people with weakened immune systems can be seriously affected by chickenpox. As a part of this project, an algorithm for the automated autonomous varicella detection has been developed and tested on various images of individuals affected by this disease. Related images were taken under different conditions which permitted to test the degree of recognition and its robustness. This technique allows to assess quantitatively any changes in the process of treatment and response to available medications used to remedy or minimize the symptoms of this disease.

Algorithmic systematized neuron cells' morphology evaluation

Alzheimers disease poses an irreversible, progressive brain disorder. It destroys affected persons memory, thinking skills, behavior and eventually the ability to carry out the simplest tasks. First symptoms of Alzheimers disease typically appear in people being in their mid-sixties. In preclinical stage of Alzheimers disease, when a person has no apparent symptoms, toxic type changes can be noticed occurring in the brain. Abnormal deposits of proteins form amyloid plaques and tau tangles throughout the brain, and once-healthy neurons stop functioning, lose connections with other neurons, and die. They manifest neurodegeneration. We propose systematized algorithm for neuron morphology evaluation in neuron images, with the attempt to distinguish healthy from degenerated neurons. This algorithm allows to derive numerical indicators representing abnormalities in neuron morphology examination.

Cardiopulmonary acoustic events classification

The acoustic cardiopulmonary signals contain complex but quantifiable essential diagnostic signs. Particularly, the lung sounds associated with inhalation and exhalation, and the heart sounds as particularly characterized by their principal components S1 and S2 when quantitatively utilized could improve the classification accuracy of diverse disease conditions. This project is related to the events classification (inhalation, exhalation, S1 and S2) with Gaussian Mixture Models (GMM) representing states in Hidden Markov Models (HMM), which were further enhanced by applying Principal Component Analysis (PCA) on Mel Frequency Cepstral Coefficients (MFCC) and Quartiles acoustic vectors. These classification outcomes drive the implementation of events quantification techniques, and also the implementation of automated auscultation for enhanced medical diagnosis. In particular, the experiments were carried out and achieved more than 88% in efficiency classification. The use of PCA showed that it could improve the outcomes classifications of inhalation-exhalation signals when noise was present, but PCA did not improve the efficiency classification of noise free S1-S2 signals. The GMM-HMM performed better applying MFCC vectors in both cases, inhalation-exhalation and S1-S2 events. The clusters analysis with Quartile vectors identifies more clusters in inhalation-exhalation signals than in S1-S2 signals. These classification outcomes drive the implementation of events modeling techniques, and also the implementation of automated auscultation with a simple laptop, which could be found in most medical offices, even in rural areas, and utilized for assisted diagnosis and documentation.

Automated nanostructure microscopic image characterization and analysis

Nanoparticles represent material particles in which one dimension measures ~100 nanometers or less. When processed into nanoparticles, the properties of many conventional organic and inorganic materials change. Since nanoparticles can be made from organic or inorganic substances, they are versatile in potential technological applications, from delicate electronics to revolutionary medical procedures. While an average structure and size is clear from characterization measurements and observations there is always a dispersity in size and shape of the final nanoparticle system. Even though microscopy is an excellent technique in capturing direct images of the nanoparticle morphology, it is difficult to assess the dispersity in size and shape of the nanostructure simply by observation of the microscopy data. This is why we propose a computer-assisted tool developed for the purpose of facilitated nanoparticle detection and its morphology identification.

Detección Automática y Clasificación de Eventos en Sonidos Cardiopulmonares de Sujetos Saludables

RESUMEN En este estudio, se presenta una metodología para evaluar lentes fáquicos intraoculares, cuando el flujo del humor La auscultación de señales basada en un estetoscopio estándar y/o electrónico no solo incluye sonidos internos del cuerpo, también incluye frecuentemente ruido externo de interferencia con componentes en el mismo rango. Esta forma de examinar es incluso afectada por los umbrales auditivos variantes de los profesionales de la salud y el grado de experiencia en reconocimiento de indicadores peculiares. Además, los resultados son a menudo caracterizados en términos cualitativos descriptivos sujetos a interpretaciones individuales. Para direccionar esta preocupación, los estudios presentados en este artículo contienen un procesamiento concurrente de las componentes dominantes de sonidos del corazón (HS) y del pulmón (HS), y una etapa de acondicionamiento que incluye la reducción de HS presente en señales LS. Específicamente, la transformada de Hilbert fue una técnica de caracterización para HS. En el caso de señales enfocadas a LS, las técnicas de detección de actividad de voz y el cálculo de umbrales de algunos componentes de los vectores acústicos de Coeficientes Cepstrales en Frecuencia Mel (MFCC), fueron útiles en la caracterización de eventos acústicos asociados. Las fases de inspiración y expiración fueron diferenciadas por medio de la sexta componente de MFCC. Con el fin de evaluar la eficiencia de esta aproximación, proponemos los Modelos Ocultos de Markov con Modelos Mesclados Gaussianos (HMM-GMM). Los resultados utilizando esta forma de detección son superiores cuando se desarrolla la clasificación con modelos HMM-GMM, la cual refleja las ventajas de la forma de detección cuantificable y clasificación sobre la aproximación clínica tradicional.

Thoracic sounds components separation integrating VAD and HT

Thoracic auscultatory signs are commonly evaluated as a part of standard diagnostic procedures which might be focused on cardiac (HS) or pulmonary (LS) origin indicators depending on evaluators aims. However, due similarities of these sounds spectral components and their extremely small intensity especially in the presence of common external noise, a separation of their origin components can be difficult even for an experienced practitioner. This is specially challenging if he/she is presented with prior recordings of these sounds as a part of related auscultatory patients record review. Differences in periodicity of these cardiac and pulmonary origin sounds assist in qualitative characterization while combined with proposed methods may lead to improved quantitative characterization and origin depended separation of their characteristics. This proposed and verified technique emphasizes initial localization of the HS related components and their subsequent extraction. Though, an examination of both HS and LS components is conducted concurrently as a part of an overall thoracic sounds analysis. In particular, as experimentally optimized, in the case of HS components assessment the Hilbert Transform (HT) is used to detect extreme points; while, in the LS components evaluation Voice Activity Detection techniques (VAD) and the calculation of thresholds of some components of acoustic vectors MFCC, are useful to detect and localize the events of particular examiners interest. Also the inspiratory and expiratory phases can be differentiated by means of the sixth component of Cepstral Coefficient of Mel Frequency (MFCC). In order to evaluate the efficiency of this technique, it was decided to apply Hidden Markov Models with Mixed Gaussian Models (HMM-GMM). The events extracted from manual and automatic detection were used to calculate HMMGMM models and perform classification. Overall, this approach allows easier detection of any anomalies and their possible origin in complex cardiopulmonary disorders.