A bio-inspired convolution neural network architecture for automatic breast cancer detection and classification using...

Area of Science:

• Bioinformatics
• Computational Biology
• Oncology

Background:

• Breast cancer is a leading global health challenge for women, necessitating early detection for effective treatment and improved survival rates.
• Gene expression data presents complexities like high dimensionality, posing challenges for accurate breast cancer detection.
• Early diagnosis is critical for identifying new treatment options, enhancing patient quality of life, and increasing survival rates.

Purpose of the Study:

• To propose a bio-inspired Convolutional Neural Network (CNN) model for enhanced breast cancer detection using gene expression data.
• To address the complexities of gene expression data, including dimensionality and bias, for improved diagnostic accuracy.
• To evaluate the performance of the proposed model against traditional CNN and other hybrid algorithms.

Main Methods:

• Utilized gene expression data from The Cancer Genome Atlas (TCGA) comprising 1208 samples (113 normal, 1095 cancerous) with 19,948 genes.
• Implemented Array-Array Intensity Correlation (AAIC) for outlier removal and normalization for data pre-processing.
• Applied gene filtration (threshold 0.25), image conversion, and grayscale transformation, followed by a hybrid CNN-Ebola Optimization Search Algorithm (EOSA) model.

Main Results:

• The proposed hybrid CNN-EOSA model achieved high classification performance: 98.3% accuracy, 99% precision, 99% recall, 99% F1-score, 90.3% kappa, 92.8% specificity, and 98.9% sensitivity for the cancerous class.
• Outperformed traditional CNN and other hybrid models including WOA-CNN, GA-CNN, SBO-CNN, LCBO-CNN, and MVO-CNN in breast cancer detection.
• Demonstrated the effectiveness of the bio-inspired approach in handling complex gene expression data for reliable detection.

Conclusions:

• The proposed bio-inspired CNN model with the Ebola Optimization Search Algorithm (EOSA) shows significant potential as a reliable and precise tool for breast cancer detection.
• The method effectively addresses challenges in gene expression data, paving the way for improved early diagnosis.
• This approach supports personalized therapy by providing accurate detection crucial for timely and effective patient management.