Predicting Breast Cancer Based on Optimized Deep Learning Approach.

Area of Science:

• Oncology
• Medical Informatics
• Artificial Intelligence

Background:

• Accurate breast cancer diagnosis is critical for effective treatment and patient outcomes.
• Machine learning (ML) and deep learning offer promising tools to aid clinicians in diagnostic decision-making.
• Existing ML models require optimization for enhanced diagnostic performance.

Purpose of the Study:

• To propose an optimized deep recurrent neural network (RNN) model for breast cancer diagnosis.
• To enhance diagnostic accuracy by optimizing RNN architecture and feature selection.
• To compare the performance of the optimized deep RNN against traditional ML models.

Main Methods:

• Developed an optimized deep RNN using Keras-Tuner, featuring optimized neuron counts and dropout rates across five hidden layers.
• Implemented three distinct feature-selection methods to identify crucial diagnostic indicators.
• Compared the optimized deep RNN against five standard ML models (Decision Tree, SVM, Random Forest, Naive Bayes, KNN) using selected features.

Main Results:

• The optimized deep RNN, utilizing features selected via the univariate method, demonstrated superior performance.
• Achieved the highest accuracy in both cross-validation (CV) and testing phases compared to all other evaluated models.
• Feature selection proved crucial for maximizing the efficacy of the deep learning model.

Conclusions:

• The optimized deep RNN presents a highly effective tool for improving breast cancer diagnosis accuracy.
• The Keras-Tuner optimization technique and univariate feature selection are key components for achieving superior performance.
• This approach holds significant potential for clinical application in breast cancer diagnostics.