Automatic Grading of Pathological Images of Prostate Using Multiwavelet Transform

Abstract

Manual histological grading of pathological tissue is a task done by pathologists to determine the level of malignancy of cancerous tissues. However, manual grading is inconsistent due to variations in a pathologist's judgments day to day and variations in judgment from pathologist to pathologist. This paper presents a new method for automatic grading of pathological images of the prostate based on the Gleason grading system. Gleason grading from very well differentiated (grade 1) to very poorly differentiated (grade 5) is usually done by viewing the low magnification microscopic image of the cancer. The lower the Gleason score, the better the patient is likely to do. According to this automated system, each cancerous specimen is assigned one of five grades based on features extracted from the multiwavelet transform of images. Specifically, energy and entropy features are extracted from submatrices obtained in decomposition, then a k-NN classifier is used to classify each image. The authors also used features extracted by wavelet packet decomposition (Daubechies wavelet D6 and D20) and second order moments to see how they compared with multiwavelet transform. The leaving-one-out technique was used to estimate error rate. One hundred graded prostate tissue sample images were processed by the automated approach. The results show the superiority of multiwavelet transform for grading compared with the other techniques. The first level of decomposition was very sensitive to noise and should not be used for feature extraction. In terms of second-level decomposition, critically sampled preprocessing was less sensitive to noise than repeated row preprocessing. The authors conclude that better classification can be achieved using second and higher levels of decompositions, while selecting the best features for classification.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Oct 25, 2001
Accession Number
ADA410854

Entities

People

  • Hamid Soltanian-Zadeh
  • Kourosh J. Khouzani

Organizations

  • University of Tehran

Tags

Communities of Interest

  • Biomedical

DTIC Thesaurus Topics

  • Cancer
  • Classification
  • Coefficients
  • Data Sets
  • Decomposition
  • Feature Extraction
  • Low Resolution
  • Medical Personnel
  • Neoplasms
  • Noise
  • Pattern Recognition
  • Preprocessing
  • Prostate
  • Prostate Cancer
  • Signal Processing
  • Test Sets
  • Two Dimensional

Readers

  • Computer Vision.
  • Image Processing and Computer Vision.
  • Trauma Surgery or Emergency Medicine.

Technology Areas

  • AI & ML
  • AI & ML - Machine Translation
  • AI & ML - Neural Networks