Abstract:
This paper presents a digital image processing software implementation to identify and study models of infection related to Chagas disease caused by Trypanosoma cruzi parasite.
The main objective is predicting disease progression to decide the appropriate treatment depending on the tissue damage observed and evaluate the effectiveness of these treatments in diminishing the structural alterations.
For this reason, digital images from colored tissue samples are collected, to be able perform the quantification of inflammatory infiltrate present in these samples.
This quantification process automation guarantees greater accuracy, objectivity obtaining results and savings in processing times of the samples involved.
The software also includes a unique image repository and metadata implementation to provide more efficiency in post-processing activities..
Palabras clave- Image processing, Chagas Image analysis, morphological operations, inflammatory infiltrate.
Referencia- Este trabajo es el resultado de un proyecto interdisciplinario
llevado a cabo por investigadores del G.In.T.E.A de la UTN-FRC
y del Laboratorio Experimental de Chagas del Instituto de Fisiología de la UNC.
Abstract:
Portable cardiac monitoring devices acquire and store all electrocardiography data of a patient for 24 hours, which are then analyzed by medical professionals. However, certain cardiac abnormalities occur sporadically
and are not recorded. In addition, it is of great importance to preprocess the data acquired in real time to filter data and generate alarms. This imposes the use of simple computational strategies for limited resources portable devices. This paper presents the development of an algorithm that allows portable
electrocardiography devices to make data storage/transmission more efficient than traditional monitoring equipment. The algorithm performs filtering and segmentation of the signal based on simple transformations and detection of peaks.
Then, the beats are decomposed into 5 characteristics by Independent Component Analysis, and classified by oneclass Support Vector Machine. The classification used is binary (normal or abnormal beats), not allowing to specify pathologies, but reducing the data. The algorithm was tested
with 20500 normal beats and 6500 abnormal beats, reducing data storage/transmission by 64 %, with 99 % accuracy. The robustness was verified on the detection and classification of the beats, and a proof of concept was performed to validate
the proposal
Palabras clave- ECG, Signal Processing, Signal Filtering, Classification, SVM.