Automatic Gleason Grading of Prostate Cancer using Gabor Filter and Local Binary Patterns

摘要

Prostate Cancer is one of the most common types of cancer found in men aged over 40 years. Detection and staging is the most critical step for pathologists. This research supports the development of Computer Aided Detection system capable of grading prostate cancer with high accuracy and less human involvement. Real patient dataset collection is a challenge, we collected real and graded dataset from Shaukat Khanum Cancer Research Hospital, Pakistan. Texture feature sets are extracted using Gabor and Local Binary Patterns with different variations. The proposed system showed an improved accuracy due to fusion of different texture features. The goal of this study is to grade the E&H stained histological images into benign, grade 3, grade 4 or grade 5. K-Nearest Neighbor classifier is used and dataset is divided randomly into training and testing using 10-fold cross validation. The proposed system shows overall accuracy of 98.3% for real dataset of 268 histological E&H images collected from 160 different patients at different times.