Integrating image processing with deep convolutional neural networks for gene selection and cancer classification...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Microarray technology enables large-scale gene expression analysis but faces challenges like high dimensionality, noise, and sparsity.
• Robust analytical methods are crucial for extracting meaningful biological insights from complex genomic data.
• Image processing aids in pattern recognition from visual data, supporting biomarker discovery in cancer research.

Purpose of the Study:

• To develop a novel framework for gene selection and cancer classification using deep neural networks and microarray data.
• To address the inherent challenges of high dimensionality, noise, and sparsity in genomic datasets.
• To enhance the accuracy and efficiency of cancer subtype prediction.

Main Methods:

• Integration of Minimum Redundancy Maximum Relevance (mRMR) for gene selection with a deep Convolutional Neural Network (CNN).
• Application of advanced preprocessing techniques and hyperparameter optimization for improved performance.
• Validation on The Cancer Genome Atlas (TCGA) and Autism Brain Imaging Data Exchange (AHBA) datasets.

Main Results:

• The Gene-Optimized Neural Framework (GONF) achieved high classification accuracy: 97% on TCGA and 95% on AHBA datasets.
• Significant reduction in false positive and false negative rates compared to existing methods.
• Demonstrated robustness and adaptability, providing biologically interpretable results.

Conclusions:

• The proposed GONF framework effectively handles high-dimensional, noisy, and sparse microarray data for cancer genomics.
• GONF offers improved accuracy and reliability in cancer subtype prediction.
• The framework shows potential for broader applications in genomic studies and clinical settings.