Browsing by Subject "bronchoscopy"

Now showing 1 - 4 of 4

Access status: Open Access ,
Generowanie danych z obrazów bronchoskopowych w celu późniejszej klasyfikacji
(Wydawnictwa AGH, 2009) Mikrut, Zbigniew; Duplaga, Mariusz
The goal of the experiments was to compare HS pixel representation with HSV (Hue Saturation Value) during detection of bleedings in bronchoscopy images. The interactive algorithm was developed to refine the bleeding regions pointed out by the doctor. Six different images were chosen and the bleeding areas were extracted based on the developed algorithm. The mutual percentage coverages of the bleeding regions were computed and compared for the two pixel representations.
Access status: Open Access ,
Śledzenie obszarów zainteresowania w sekwencjach obrazów bronchoskopowych za pomocą metody SIFT
(Wydawnictwa AGH, 2009) Pawlik, Piotr; Bubliński, Zbigniew; Duplaga, Mariusz
The paper presents an attempt to utilize the modified SIFT method for ROI (region of interest) tracking on sequences of bronchoscopy images. The method was tested on several movies recorded during medical treatment and very promising results were obtained. Also, the areas of possible improvements were pointed out.
Access status: Open Access ,
Szybka detekcja rozgałęzień w drzewie oskrzelowym
(Wydawnictwa AGH, 2010) Pawlik, Piotr; Bubliński, Zbigniew
The paper presents an algorithm for fast detection of bronchial tree subdivisions on images obtained during bronchoscopy examination. Localization of these subdivisions could be very helpful to physicians while reviewing collected data. The time of the detection should be as short as possible, especially when large sets of data were gathered during examination. The presented algorithm works in real time and is characterized by a very high level of specificity.
Access status: Open Access ,
Wykrywanie krwi na obrazach bronchoskopowych za pomocą sieci neuronowych
(Wydawnictwa AGH, 2009) Mikrut, Zbigniew; Duplaga, Mariusz
In the paper the experiments with using SOM-supervised neural networks for pixel (HSV) classification were presented. Six visually different images were chosen to be the basis for the SOM training. For these images learning sets were created based on the refined masks of the bleeding regions pointed out by the doctor. Next the six learning sets were merged and the ambiguous pixel representations were removed. Two types of SOM-supervised networks (of »normal« and »small« sizes) were created and learned. The classification results were obtained and analyzed both for learning sets and for 14 test images. Several conclusions were stated concerning the learning methodology and the bleeding areas postprocessing.