Computed Tomography Image under Convolutional Neural Network Deep Learning Algorithm in Pulmonary Nodule Detection...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• Pulmonary nodules require accurate detection and segmentation for effective diagnosis and treatment planning.
• Computed Tomography (CT) is a primary imaging modality for pulmonary nodule assessment.
• Current segmentation and detection methods may have limitations in accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate a deep learning-based model for the segmentation and extraction of pulmonary nodules from CT images.
• To compare the performance of Mask-RCNN and R-FCN algorithms in pulmonary nodule detection and diagnosis.
• To assess the diagnostic accuracy and efficiency improvements offered by the proposed algorithm.

Main Methods:

• A Mask-RCNN algorithm model was developed for end-to-end image segmentation of pulmonary nodules.
• The R-FCN structure was employed within the Mask-RCNN model for nodule detection.
• A dataset of 56 patients with pulmonary nodules was used for training and validation, with 3D Faster-RCNN used for nodule labeling.

Main Results:

• The trained Mask-RCNN model effectively segmented lung CT images, with minor boundary jitter.
• The R-FCN algorithm achieved a detection speed of 0.172 seconds/picture with 88.9% accuracy.
• The deep learning algorithm demonstrated high diagnostic accuracy: 93.75% for malignant nodules, 91.67% for benign lesions, and 92.85% overall.

Conclusions:

• Deep learning models, particularly Mask-RCNN, can effectively segment pulmonary nodules in CT images.
• The proposed approach significantly improves the diagnostic efficiency and accuracy of pulmonary nodule detection.
• This study provides a strong theoretical basis for advancing pulmonary nodule diagnosis and lung cancer treatment.