Analysis of Machine Learning Algorithms for Diagnosis of Diffuse Lung Diseases

Area of Science:

• Medical Image Processing
• Computational Intelligence
• Biomedical Image Registration

Background:

• Diffuse lung diseases (DLDs) present diagnostic challenges due to their diversity and unknown causes.
• Computer-aided diagnosis (CAD) using machine learning (ML) can improve diagnostic accuracy for DLDs.

Purpose of the Study:

• To explore dimensionality reduction combined with ML for DLD diagnosis.
• To enhance classification accuracy beyond state-of-the-art methods.

Main Methods:

• Utilized a dataset of 3252 regions of interest (ROIs) with 28 features per ROI.
• Applied Principal Component Analysis, Linear Discriminant Analysis, and Stepwise Selection for feature reduction.
• Employed Support Vector Machine, Gaussian Mixture Model, k-Nearest Neighbor, Deep Feedforward Neural Network, and Deep Convolutional Neural Network for classification.

Main Results:

• Achieved maximum dimensionality reduction from 28 to 5 features using Linear Discriminant Analysis (LDA).
• The Deep Feedforward Neural Network (DFNN) achieved the highest overall classification accuracy at 99.60%.

Conclusions:

• This research contributes to identifying effective features for characterizing diffuse lung diseases.
• The study highlights the potential of optimized feature selection in improving DLD diagnosis through computational intelligence.