Sistem diagnosis awal barah pangkal rahim berasaskan rangkaian neural
Loading...
Date
2002-08
Authors
Mat Isa, Norashidi
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Pap test is an effective screening test for cervical cancer. However, Pap test has three
disadvantages namely it does not always produce accurate diagnosis result due to
human and technical errors, requires lot of pathologists services and diagnosis result can
only be obtained after a long duration. Therefore, the current study focused on
developing a diagnosis support system for Pap test. In image processing system,
automatic seed based region growing (ASBRG) algorithm is proposed to segment the
Pap smear image into nucleus, cytoplasm and background areas. The proposed
automatic algorithm produced better segmentation performance as compared to
conventional techniques by ensuring complete segmentation process and maintaining
the original size and shape of cell structures. The current study also proposed Pap smear
image segmentation technique based on moving k-means clustering. The proposed
technique was capable to reduce dead cluster, cluster redundancy and trapped cluster in
local minima problem. It gave better performance as compared to conventional
clustering techniques. The current study also proposed region growing based feature
extraction (RGBFE) algorithm to extract size and grey level of nucleus and cytoplasm
automatically. The suggested combination of moving k-means and linear contrast was
shown to be capable of enhancing the contrast of Pap smear image for easier screening
process by pathologists. In proposed diagnosis system, the current study go one step
further as compared to previous studies by classifying cervical cell into normal, LSIL
and HSIL cell. For that purpose, hierarchical neural network has been proposed. The
results. proved that the hierarchical hybrid multilayered perceptron (H2MLP) network
produced better diagnostic performance as compared to otter hierarchical, hybrid and
conventional neural networks. The current study has also successfully proposed two
new analysis called diagnosis confidence level and confidence percentage to give
clearer picture on the confidence level for each diagnosis. Based on these researches, an
early diagnosis system for cervical cancer based on neural network has been developed
by. ...~ om.. bining jhe ima. ge processing system .a nd . diagnosis system. The prop.csed system
is called NeuralPap. NeuralPap is an automatic diagnosis system. First, NeuralPap will
segment the Pap smear image by using ASBRG algorithm. Then, the cervical cell
features will be extracted using RGBFE algorithm. Lastly, by using the extracted
features, the diagnosis will be carried out by H2MLP network. For overall diagnostic
performance, NeuralPap produced 94.29%, 89.42% and 100.00% of accuracy,
sensitivity and specificity respectively. On top of that, NeuralPap also provides
supplementary facilities which are image processing, diagnostic performance
comparison using various neural networks and general information about neural
network and cervical cancer.
Description
Keywords
Barah pangkal rahim , Rangkaian neural