Fast opposite weight learning rules with application in breast cancer diagnosis

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Machine Learning

Background:

• Breast mass classification is challenging for radiologists.
• Computer-aided diagnosis (CADx) systems can assist radiologists.
• Neural Networks (NN), like Multilayer Perceptrons (MLP), offer high accuracy but suffer from long training times.

Purpose of the Study:

• To introduce novel learning rules (OWBPP, OWBPE, OWBPI) to accelerate MLP training.
• To develop and evaluate CADx systems for breast mass diagnosis using these novel algorithms.
• To quantitatively analyze the accuracy and convergence rates of different MLP training algorithms.

Main Methods:

• Development of three novel learning rules: Opposite Weight Back Propagation per Pattern (OWBPP), per Epoch (OWBPE), and per Pattern in Initialization (OWBPI).
• Implementation of MLP classifiers trained with traditional Back Propagation (BP) and the novel algorithms.
• Quantitative analysis of classifier accuracy and convergence rate using Receiver Operating Characteristic (ROC) analysis.

Main Results:

• The OWBPE algorithm demonstrated a convergence rate over 4 times faster than traditional BP.
• CADx systems utilizing the OWBPE classifier achieved an average area under the ROC curve (Az) of 0.928.
• The OWBPE-based CADx systems reported a False Negative Rate (FNR) of 9.9% and a False Positive Rate (FPR) of 11.94%.

Conclusions:

• The proposed OWBPE learning rule significantly accelerates MLP training for CADx systems.
• The OWBPE algorithm enhances the diagnostic performance of CADx systems for breast mass classification.
• Novel learning rules offer a promising approach to improve the efficiency and accuracy of medical diagnostic tools.