Deep learning-based algorithm for lung cancer detection on chest radiographs using the segmentation method.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Radiology

Background:

• Lung cancer is a leading cause of mortality worldwide.
• Early detection of lung cancer significantly improves patient outcomes.
• Chest radiography remains a primary screening tool, but interpretation can be challenging.

Purpose of the Study:

• To develop and validate a deep learning (DL)-based model for lung cancer detection on chest radiographs.
• To assess the model's sensitivity and false positive rate in an independent test dataset.
• To evaluate the model's performance in detecting lung cancers in various anatomical locations.

Main Methods:

• A DL model was developed and trained using a dataset of 629 chest radiographs with 652 nodules/masses.
• Model validation was performed using five-fold cross-validation.
• The model's performance was assessed on an independent test set of 151 radiographs with 159 nodules/masses, evaluating sensitivity and mean false positive indications per image (mFPI).
• The Dice coefficient was calculated for malignant lesions.

Main Results:

• The DL model achieved a sensitivity of 0.73 and an mFPI of 0.13 on the test dataset.
• Sensitivity was lower for lung cancers overlapping with anatomical blind spots (0.50-0.64) compared to non-overlapped locations (0.87).
• The average Dice coefficient for malignant lesions was 0.52.

Conclusions:

• The developed DL-based model demonstrates capability in detecting lung cancers on chest radiographs.
• The model exhibits a low rate of false positives, indicating potential for clinical utility.
• Further refinement is needed to improve detection in challenging anatomical regions, such as pulmonary apices and the cardiac silhouette.