Lung cancer prediction using neural network ensemble with histogram of oriented gradient genomic features

Area of Science:

• Bioinformatics
• Computational Biology
• Genomic Medicine

Background:

• Lung cancer prediction remains a challenge, necessitating advanced computational methods for early detection.
• Genomic biomarkers, including mutations in EGFR, KRAS, and TP53, are crucial for understanding lung cancer development.
• Machine learning offers potential for analyzing complex genomic data for diagnostic purposes.

Purpose of the Study:

• To experimentally compare artificial neural network (ANN) and support vector machine (SVM) ensembles against their nonensemble variants for lung cancer prediction.
• To evaluate the efficacy of different feature extraction methods in conjunction with these classifiers.
• To assess the potential of machine learning models for automated lung cancer screening.

Main Methods:

• Utilized the IGDB.NSCLC corpus containing patient nucleotide samples with specific genomic mutations.
• Employed Voss DNA encoding to convert nucleotide sequences into numerical genomic sequences.
• Applied Histogram of Oriented Gradient (HOG) and Local Binary Pattern (LBP) for feature extraction from encoded genomic data.
• Trained and compared ANN and SVM classifiers, both in ensemble and nonensemble configurations.

Main Results:

• The artificial neural network (ANN) ensemble combined with Histogram of Oriented Gradient (HOG) features achieved the highest accuracy (95.90%) and lowest mean square error (0.0159) on the training dataset.
• This combination demonstrated superior performance compared to other tested machine learning models and feature extraction techniques.
• The results indicate a strong correlation between genomic features and lung cancer presence.

Conclusions:

• The ANN ensemble utilizing HOG genomic features presents a promising approach for the automated screening and early detection of lung cancer.
• This methodology can potentially assist pathologists in identifying patients who would benefit from targeted molecular therapies.
• The findings support the integration of advanced machine learning techniques into clinical practice for improved lung cancer management.