Research on lung cancer diagnosis based on machine learning

Area of Science:

• Computational biology
• Medical informatics
• Machine learning in oncology

Background:

• Accurate tumor classification is crucial for effective cancer treatment.
• Differentiating benign from malignant tumors presents a significant diagnostic challenge.
• Machine learning offers potential solutions for improving diagnostic accuracy and speed.

Purpose of the Study:

• To evaluate machine learning methods for classifying lung tissue as benign or malignant.
• To assess the performance of various algorithms in determining lung cancer aggressiveness.
• To compare methods with and without built-in feature selection for lung cancer diagnosis.

Main Methods:

• Employed artificial neural networks (ANN) and logistic regression variants.
• Utilized methods without built-in feature selection: ANN, logistic regression, ridge penalized logistic regression.
• Applied methods with built-in feature selection: lasso penalized logistic regression, elastic-net penalized logistic regression, and hybrid L1/2 + L2 regularization (HLR).

Main Results:

• ANN achieved 91.82% accuracy in classifying benign/malignant lung tissue (without feature selection).
• HLR achieved 96.67% accuracy in classifying benign/malignant lung tissue (with feature selection).
• ANN attained 84.74% accuracy in assessing lung cancer aggressiveness (without feature selection).
• HLR reached 93.33% accuracy in assessing lung cancer aggressiveness (with feature selection).

Conclusions:

• HLR demonstrated superior performance in classifying lung tissue and assessing cancer aggressiveness when feature selection is integrated.
• ANN showed strong competitiveness among methods without built-in feature selection for both classification tasks.
• Both HLR and ANN show significant potential for improving lung cancer diagnosis and grading.