Classification of Breast Cancer Images by Implementing Improved DCNN with Artificial Fish School Model.

Area of Science:

• Biomedical Imaging
• Artificial Intelligence in Medicine
• Computational Pathology

Background:

• Breast cancer diagnosis relies on pathological images, with existing methods like radial basis neural networks (RBF) often separating feature extraction and reduction.
• Conventional techniques can be time-consuming, highlighting the need for more efficient diagnostic approaches.

Purpose of the Study:

• To develop an optimized deep convolutional neural network (CNN) model for accurate and efficient breast cancer image classification.
• To integrate feature extraction, reduction, and classification into a single CNN process, overcoming limitations of existing RBF techniques.

Main Methods:

• A novel CNN model was developed, incorporating an artificial fish school algorithm for optimization.
• The model utilized a wiener filter for preprocessing and direct training data input to the deep CNN via the optimization algorithm.
• Feature extraction, reduction, and classification were performed within a single deep CNN framework.

Main Results:

• The proposed model demonstrated improved performance in classifying normal, benign, and malignant breast tissues compared to existing RBF techniques.
• The artificial fish school optimization algorithm reduced error rates and increased performance efficiency by optimizing epochs and training data for the deep CNN.
• Key performance metrics including sensitivity, accuracy, specificity, F1 score, and recall were enhanced.

Conclusions:

• The integrated deep CNN approach with artificial fish school optimization offers a more efficient and accurate method for breast cancer diagnosis from pathological images.
• This method streamlines the diagnostic process by combining multiple steps into a single, optimized network, potentially reducing diagnostic time and improving patient outcomes.