Automated lung cancer diagnosis using three-dimensional convolutional neural networks

Area of Science:

• Medical Imaging
• Oncology
• Computer-Aided Diagnosis

Background:

• Lung cancer is a leading cause of cancer-related mortality globally.
• Early detection via low-dose computed tomography (LDCT) significantly improves patient survival rates.
• Accurate identification and characterization of lung nodules are critical for timely intervention.

Purpose of the Study:

• To develop and validate a general framework for lung cancer detection in LDCT images.
• To enhance the accuracy of lung nodule detection and malignancy prediction.
• To establish a system that achieved top performance in a competitive challenge.

Main Methods:

• A two-stage approach involving nodule detection and cancer prediction.
• Nodule detection trained on the LIDC-IDRI dataset.
• Cancer prediction trained on the Kaggle DSB 2017 dataset and evaluated on the ISBI 2018 challenge dataset.

Main Results:

• Candidate extraction achieved a recall of 99.6%.
• False positive reduction increased precision by a factor of 2000.
• The cancer predictor achieved a ROC AUC of 0.913, ranking 1st in the ISBI 2018 challenge.

Conclusions:

• The proposed framework demonstrates high efficacy in detecting lung nodules and predicting malignancy from LDCT scans.
• The system's performance highlights the potential of advanced computational methods in improving lung cancer diagnosis.
• This approach offers a promising tool for early lung cancer detection and management.