Multi-scale gradual integration CNN for false positive reduction in pulmonary nodule detection

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer-Aided Diagnosis

Background:

• Early lung cancer detection is critical for reducing mortality.
• Computer-aided systems for pulmonary nodule detection on CT scans are of great interest.
• Accurate candidate nodule detection is challenging due to morphological variations and organ mimicry.

Purpose of the Study:

• To propose a novel Multi-scale Gradual Integration Convolutional Neural Network (MGI-CNN) for improved pulmonary nodule detection.
• To address the challenges in candidate nodule detection by leveraging multi-scale contextual information and feature integration.

Main Methods:

• The MGI-CNN utilizes multi-scale inputs with varying contextual information.
• It employs gradual integration of abstract information from different input scales.
• The network learns multi-stream feature integration in an end-to-end manner.

Main Results:

• Experiments were conducted on the LUNA16 challenge datasets.
• The MGI-CNN achieved an average candidate nodule detection performance (CPM) of 0.908 on V1 and 0.942 on V2.
• The proposed method significantly outperformed state-of-the-art methods.

Conclusions:

• The MGI-CNN demonstrates superior performance in pulmonary nodule detection.
• This novel network offers a promising advancement for early lung cancer diagnosis using CT scans.
• The implementation is available in Python using TensorFlow.