Hybrid models for lung nodule malignancy prediction utilizing convolutional neural network ensembles and clinical...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lung cancer has high incidence and mortality rates globally.
• Early detection of precancerous lesions is crucial for improved patient outcomes.
• Low-dose computed tomography (LDCT) is a key tool for lung cancer screening and diagnosis.

Purpose of the Study:

• To enhance lung nodule classification and malignancy prediction using convolutional neural networks (CNNs).
• To investigate the combined impact of nodule size and clinical information on cancer risk assessment.
• To develop a hybrid CNN model integrating size and clinical data for improved diagnostic accuracy.

Main Methods:

• Utilized a dataset from the National Lung Screening Trial.
• Categorized lung nodules into large and small groups based on clinical guidelines.
• Developed and trained individual CNN models for each group (size and clinical features).
• Created an ensemble hybrid model combining CNNs for size and clinical data.

Main Results:

• Achieved an Area Under the Curve (AUC) of 0.9.
• Obtained an accuracy of 83.12%, representing a significant improvement.
• Demonstrated that integrating nodule size and clinical information enhances malignancy prediction.

Conclusions:

• Segmenting nodules by size and incorporating clinical data improves predictive model performance.
• A hybrid approach combining clinical information and size groups further refines lung cancer risk prediction.
• This study highlights the potential of AI in early lung cancer detection and risk stratification.