Investigation on mammographic image compression and analysis using multiwavelets and neural network

In digital mammography, the resulting electronic image is very large in size. Hence, the size poses a big challenge to the transmission, storage and manipulation of images. Microcalcification is one of the earliest sign of breast cancer and it appears in small size, low contrast radiopacites in high...

Full description

Saved in:
Bibliographic Details
Main Authors: Ragupathy, U. S., Kumar, A. S.
Format: Conference Proceeding
Language:English
Subjects:
Feature extraction
Hospitals
Image coding
Image Compression
Image reconstruction
Mammography
Microcalcification
Multiwavelet
Neural Network
Neurons
PSNR
Wavelet transforms
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In digital mammography, the resulting electronic image is very large in size. Hence, the size poses a big challenge to the transmission, storage and manipulation of images. Microcalcification is one of the earliest sign of breast cancer and it appears in small size, low contrast radiopacites in high frequency spectrum of mammographic image. Scalar wavelets excel multiwavelets in terms of Peak Signal - to Noise Ratio (PSNR), but fail to capture high frequency information. Multiwavelet preserves high frequency information. This paper proposes multiwavelet based mammographic image compression, and microcalcification analysis in compressed reconstructed images against original images using multiwavelets and neural networks. For a set of four mammography images, the proposed balanced multiwavelet based compression method achieves an average PSNR of 9.064 dB greater than the existing compression scheme. It also details the classification results obtained through the multiwavelet based scheme in comparison with the existing scalar wavelet based scheme. For a testing sample of 30 images, the proposed classification scheme outperforms the scalar wavelet based classification by sensitivity of 2.23% and specificity of 3.4% for original (uncompressed) images. Also it increases the sensitivity by 2.72% and specificity by 8.4% for compressed reconstructed images. This increase in sensitivity and specificity reveals a better performance of the proposed detection scheme.
DOI:10.1109/ICoBE.2012.6178947