A Method for Real-Time Lung Nodule Instance Segmentation Using Deep Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lung cancer screening is vital for early detection, reducing mortality by 20-30% in high-risk groups.
• Deep learning, particularly computer vision, excels at object detection in images.
• Accurate segmentation of lung masses is crucial for diagnosis and treatment planning.

Purpose of the Study:

• To develop and evaluate a deep learning method for lung mass instance segmentation and classification.
• To improve the accuracy and efficiency of lung cancer screening through automated analysis.
• To classify detected lung masses as nodules, cancer, or adenocarcinoma.

Main Methods:

• Utilized the You-Only-Look-Once (YOLO) model for lung nodule segmentation.
• Applied instance segmentation to generate masks for individual lung masses.
• Trained and tested the method on a dataset of real-world lung computed tomography (CT) images.

Main Results:

• Achieved an average precision of 0.757 and recall of 0.738 in mass classification.
• Obtained an average mask precision of 0.75 and mask recall of 0.733 for segmentation.
• Demonstrated effectiveness in detecting, classifying, and segmenting lung masses, including small ones.

Conclusions:

• The proposed deep learning method effectively performs instance segmentation and classification of lung masses.
• This technology can assist radiologists in automatic lung screening and the detection of subtle masses.
• The method shows promise for enhancing early lung cancer diagnosis and management.