Premium
Computer aided diagnosis of prostate cancer: A texton based approach
Author(s) -
Rampun Andrik,
Tiddeman Bernie,
Zwiggelaar Reyer,
Malcolm Paul
Publication year - 2016
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.4962031
Subject(s) - artificial intelligence , pattern recognition (psychology) , computer science , random forest , feature extraction , computer aided diagnosis , support vector machine , feature (linguistics) , philosophy , linguistics
Purpose In this paper the authors propose a texton based prostate computer aided diagnosis approach which bypasses the typical feature extraction process such as filtering and convolution which can be computationally expensive. The study focuses the peripheral zone because 75% of prostate cancers start within this region and the majority of prostate cancers arising within this region are more aggressive than those arising in the transitional zone. Methods For the model development, square patches were extracted at random locations from malignant and benign regions. Subsequently, extracted patches were aggregated and clustered using k ‐means clustering to generate textons that represent both regions. All textons together form a texton dictionary, which was used to construct a texton map for every peripheral zone in the training images. Based on the texton map, histogram models for each malignant and benign tissue samples were constructed and used as a feature vector to train our classifiers. In the testing phase, four machine learning algorithms were employed to classify each unknown sample tissue based on its corresponding feature vector. Results The proposed method was tested on 418 T2‐W MR images taken from 45 patients. Evaluation results show that the best three classifiers were Bayesian network ( A z = 92.8% ± 5.9%), random forest (89.5% ± 7.1%), and k ‐NN (86.9% ± 7.5%). These results are comparable to the state‐of‐the‐art in the literature. Conclusions The authors have developed a prostate computer aided diagnosis method based on textons using a single modality of T2‐W MRI without the need for the typical feature extraction methods, such as filtering and convolution. The proposed method could form a solid basis for a multimodality magnetic resonance imaging based systems.