Computer-aided diagnosis: a neural-network-based approach to lung nodule detection

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology
• Oncology

Background:

• Early detection of lung cancer nodules is crucial for improving patient outcomes.
• Computer-aided diagnosis (CAD) systems offer potential for enhancing the accuracy and efficiency of radiological interpretations.
• Artificial Neural Networks (ANNs) have shown promise in image analysis tasks, including medical image interpretation.

Purpose of the Study:

• To develop and evaluate a novel two-level artificial neural network (ANN) architecture for detecting lung cancer nodules.
• To assess the system's performance using digitized chest radiographs with both real and simulated nodules.
• To quantify the sensitivity and false positive rates of the proposed CAD system.

Main Methods:

• A two-level ANN architecture was designed for lung nodule detection.
• The first ANN identifies suspicious regions in low-resolution radiographs.
• The second ANN analyzes curvature peaks within suspicious regions to detect small tumors, leveraging their unique signature in curvature-peak feature space.

Main Results:

• The developed system demonstrated high sensitivity, ranging from 89% to 96%, depending on nodule size.
• The system achieved a low mean false positive rate of 5-7 false positives per image.
• Performance was evaluated using 60 radiographs containing 90 real and 288 simulated nodules.

Conclusions:

• The proposed two-level ANN system is effective for detecting lung cancer nodules on digitized chest radiographs.
• The system's performance, characterized by high sensitivity and a low false positive rate, supports its potential clinical utility.
• The curvature-peak feature analysis within the ANN architecture contributes to the accurate identification of small lung tumors.