Incremental learning with SVM for multimodal classification of prostatic adenocarcinoma

Area of Science:

• Medical Imaging
• Machine Learning
• Oncology

Background:

• Prostate cancer detection in MR images is challenging due to variant presentations.
• Current methods require robust pattern recognition systems for improved accuracy.

Purpose of the Study:

• To develop and evaluate a pattern recognition system for robust prostate cancer detection using multimodal MR images.
• To enhance diagnostic accuracy and efficiency for radiologists.

Main Methods:

• Utilized an incremental learning ensemble algorithm with support vector machines (SVM).
• Integrated anatomic, texture, and functional features, with texture quantified by statistical approaches and rotation invariant local phase quantization (RI-LPQ).
• Employed B-Spline interpolation, bias field correction, and intensity standardization for data preprocessing.

Main Results:

• Achieved an average sensitivity of 0.844 ± 0.068 and specificity of 0.780 ± 0.038.
• Demonstrated superior or similar performance to state-of-the-art methods on a small dataset.
• Showcased the feasibility of ensemble SVM for incremental learning and knowledge preservation.

Conclusions:

• Texture descriptors offer more discriminative patterns than functional information for prostate cancer detection.
• The system improves classification efficiency, robustness, and information selection.
• Generated probability maps aid radiologists in reducing diagnostic variability and false negatives.