Semantic Multilayer SVM: Novel Artificial Intelligence Applied to Prostate Cancer Grading and Breast Cancer Diagnosis

Overview

Objectives/Goals We develop a novel Support Vector Machine (SVM) learning algorithm called Semantic Multilayer Support Vector Machine (SMLSVM) that can take groups of low-level numerical data and translate them into higher-level concepts. We also develop an image analysis toolkit, Procist, to translate a prostate histology image into numerical data. We apply Procist and SMLSVM to prostate cancer grading based on two Gleason Grades, a novel development. Methods/Materials SMLSVM generates fewer, more valuable higher-level features out of many, lower-level numerical features. These capabilities allow it to reduce the curse of dimensionality and the semantic gap. Procist is a computer vision algorithm that can generate numerical data that characterize the spatial arrangement and irregularity of cells, nuclei, stroma, lumens, and other tissue elements. All algorithms were developed in MATLAB, and were tested on 20 prostate cancer histology images from the Johns Hopkins Medical Institute (JHMI) without patient info. We further validated SMLSVM alone on the University of Wisconsin Breast Cancer Dataset to differentiate between 699 malignant and benign tumor cases. Results We performed 10-fold cross validation on both datasets, and preliminary testing indicates that my algorithms outperformed current SVM approaches on both datasets. Results from further testing will be updated and presented during the fair. Conclusions/Discussion The objectives of this project have been achieved. SMLSVM and Procist could classify cancer malignancy and cancer severities with very high accuracies. Few studies have been performed on Multilayer SVMs, and we present new research into SVM technology, artificial intelligence, and deep learning. Furthermore, by differentiating cancers based on two Gleason Grades, we present novel contributions to computational histopathology, where current approaches only differentiate between one Gleason Grade and are much less clinically accurate. A provisional patent is pending on SMLSVM. I am currently working with JHMI to secure a much larger dataset for a better validation of my algorithms' capabilities.

Competition history

• CSEF 2014 Mathematics & Software · Entry S1408

Resources

• CSEF project summary ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Computational Sciences projects
• CSEF Computational Sciences projects
• Every CSEF 2014 project