Genetically Optimized Modular Neural Networks for Precision Lung Cancer Diagnosis: Exploratory Study of Novel...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lung cancer is a leading cause of cancer mortality.
• Low-dose computed tomography (CT) screening enhances survival but faces radiologist shortages.
• Developing efficient AI for lung cancer detection is crucial.

Purpose of the Study:

• To develop and evaluate a precise and efficient AI algorithm for lung cancer diagnosis using chest CT scans.
• To address the challenges posed by radiologist shortages in lung cancer screening.

Main Methods:

• Utilized 156 patient chest CT scans across two databases.
• Performed feature extraction using statistics, histograms, Fast Fourier Transform (FFT), Discrete Cosine Transform (DCT), and Walsh-Hadamard Transform (WHT).
• Optimized classifiers including Multi Layer Perceptron (MLP), Generalized Feed Forward Neural Network (GFF-NN), Modular Neural Network (MNN), and Support Vector Machine (SVM) using genetic algorithms.

Main Results:

• The Modular Neural Network (MNN) with FFT features and momentum learning achieved 100% classification accuracy.
• Perfect average classification accuracy was consistently obtained across both datasets during cross-validation.
• Evaluation metrics included classification accuracy, Mean Squared Error (MSE), and Area under the ROC curve (AUC).

Conclusions:

• The genetically optimized MNN classifier demonstrates high performance for lung cancer diagnosis from CT images.
• Achieving perfect classification accuracy indicates strong potential for clinical application.
• The AI tool can enhance diagnostic precision, serve as a triage system, and reduce radiologist workload.