Improved lung nodule diagnosis accuracy using lung CT images with uncertain class

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lung cancer is a leading cause of cancer mortality.
• Pulmonary nodules on CT scans are early indicators of lung cancer.
• Advancements in medical imaging increase the detection of small pulmonary nodules.

Purpose of the Study:

• To propose a novel computer-aided diagnosis (CAD) system for pulmonary nodules.
• To enhance the accuracy of lung cancer diagnosis using semi-supervised machine learning.
• To address the challenge of utilizing unlabeled data in nodule diagnosis.

Main Methods:

• Developed a feature model for pulmonary nodules from lung CT images.
• Compared Extreme Learning Machine (ELM) with Support Vector Machine (SVM), Probabilistic Neural Network (PNN), and Multilayer Perceptron (MLP).
• Proposed a semi-supervised ELM (SS-ELM) algorithm for CAD, incorporating both labeled and unlabeled data.

Main Results:

• Extreme Learning Machine (ELM) demonstrated superior performance over SVM, PNN, and MLP in training time and testing accuracy.
• The proposed SS-ELM based CAD system achieved improved testing accuracy compared to the standard ELM.
• Experiments utilized 1018 thoracic CT image sets from the LIDC-IDRI database.

Conclusions:

• The SS-ELM based pulmonary nodule CAD system offers better generalization performance.
• The system achieves faster learning speeds and higher testing accuracy than traditional methods (ELM, SVM, PNN, MLP).
• SS-ELM effectively utilizes uncertain class data for improved pulmonary nodule diagnosis.