Predicting Nodule Malignancy using a CNN Ensemble Approach.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Oncology
• Radiology

Background:

• Lung cancer is a leading global cause of cancer mortality, necessitating early detection and diagnosis.
• Computed tomography (CT) is the primary imaging modality for lung cancer screening and diagnosis.
• Radiomics and deep learning, particularly convolutional neural networks (CNNs), show promise for analyzing lung nodules.

Purpose of the Study:

• To enhance the prediction of lung nodule malignancy using an ensemble of classifiers.
• To investigate the combined predictive power of radiomics features and CNNs for lung nodule analysis.
• To improve non-invasive nodule tracking and diagnosis in low-dose CT screening.

Main Methods:

• Utilized radiomics features and both custom-trained and pre-trained CNNs for predictive analysis.
• Employed an ensemble approach, combining probability predictions from different models on an unseen test set.
• Applied the methodology to participant subsets from the National Lung Screening Trial.

Main Results:

• Ensemble models demonstrated increased accuracy and area under the receiver operating characteristic curve (AUC) compared to individual models.
• Achieved a peak AUC of 0.96 and accuracy of 89.45% with the ensemble approach.
• These results represent significant improvements over previous benchmarks (AUC 0.87, accuracy 76.79%).

Conclusions:

• Ensemble classifiers integrating radiomics and CNNs significantly enhance the accuracy of lung nodule malignancy prediction.
• This approach offers a precise, non-invasive method for nodule tracking and diagnosis, particularly in low-dose CT screening.
• The findings support the use of combined radiomics and deep learning for improved early lung cancer detection and patient outcomes.