Performance analyses of five neural network classifiers on nodule classification in lung CT images using WEKA: a...

Area of Science:

• Medical Imaging
• Machine Learning
• Computer-Aided Diagnosis

Background:

• Lung nodules are critical indicators of lung disease, necessitating accurate detection methods.
• Computer-aided diagnosis systems leverage machine learning for improved lung nodule identification.

Purpose of the Study:

• To compare the performance of five distinct neural network classifiers for lung nodule detection.
• To evaluate the efficacy of the WEKA interface for analyzing these classifiers on lung CT images.

Main Methods:

• Utilized 624 handcrafted features from 52 lung CT images from the Lung Image Database Consortium (LIDC).
• Employed WEKA software to train and validate Multilayer Perceptron (MLP), DL4JMLP, logistic regression, SGD, and simple logistic classifiers.
• Assessed classifier performance using 11 metrics, including accuracy, ROC area, and sensitivity, with tenfold cross-validation.

Main Results:

• Stochastic Gradient Descent (SGD) classifier achieved the highest accuracy (94.44%) and ROC area (0.91).
• Simple logistic classifier also showed strong performance with 88.88% accuracy and 0.93 ROC area.
• MLP classifier achieved 86.53% accuracy and 0.91 ROC area.

Conclusions:

• SGD and simple logistic classifiers are highly effective for lung nodule detection using WEKA.
• This study highlights the potential of WEKA for comparative analysis of machine learning models in medical imaging.