Computerized detection of lung nodules through radiomics

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Lung cancer remains a leading cause of cancer mortality, with survival rates dramatically lower in late stages (Stage IV, 2% 5-year survival) compared to early stages (Stage I, 50% 5-year survival).
• Early detection of lung cancer through spiral computed tomography (CT) scans is crucial for improving patient outcomes.
• Radiologists face significant workloads and potential for human error in interpreting numerous CT scans for pulmonary nodules.

Purpose of the Study:

• To propose and evaluate a computer-aided detection (CAD) system integrated with radiomics for the automated detection of pulmonary nodules.
• To enhance the accuracy and efficiency of lung cancer diagnosis in high-risk patients.
• To reduce the burden on radiologists and minimize diagnostic errors in lung cancer screening.

Main Methods:

• A nodular enhancement filter was employed for segmenting nodule candidates and extracting radiomic features.
• Synthetic minority over-sampling technique (SMOTE) was used to address class imbalance in the dataset.
• A random forest classifier was utilized to differentiate between true nodules and false positives, leveraging radiomics for heterogeneity quantification.

Main Results:

• The proposed CAD system achieved 88.9% sensitivity with an average of four false positive detections per CT scan.
• Performance was evaluated on 1004 cases from the Lung Image Database Consortium (LIDC), with 502 cases for training and 502 for testing.
• The radiomics approach effectively quantified intratumor heterogeneity and multifrequency information, correlating strongly with lung nodules.

Conclusions:

• The developed computer-aided detection system demonstrated high performance on the LIDC database.
• The proposed scheme shows potential effectiveness across various CT configurations used in routine lung cancer diagnosis and screening.
• This radiomics-based CAD system offers a promising tool for improving early lung cancer detection.