Hyperparameter optimization and development of an advanced CNN-based technique for lung nodule assessment.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Early lung cancer detection is crucial for improving patient outcomes.
• Computed tomography (CT) scans are vital for identifying lung nodules.
• Automated analysis of CT images can enhance diagnostic accuracy and efficiency.

Purpose of the Study:

• To propose an advanced methodology for automated lung nodule assessment using CT images.
• To improve early-stage lung cancer detection through enhanced nodule identification and classification.
• To develop a robust system for analyzing diverse thoracic CT datasets.

Main Methods:

• Utilized a 13-layer convolutional neural network (CNN) with a 3x3 kernel for feature extraction.
• Incorporated transfer learning with EfficientNetV2 (TLEV2N) to boost training performance.
• Employed CNN architecture for accurate benign and malignant lung nodule classification.

Main Results:

• Achieved 97.56% accuracy and 98.4% specificity in lung nodule detection.
• Significantly reduced the false-negative rate in lung cancer diagnosis.
• Demonstrated the 3x3 kernel's effectiveness in capturing pixel variations and morphological features.

Conclusions:

• The proposed non-invasive technique shows significant potential for early lung cancer detection using low-dose CT scans.
• The methodology offers improved accuracy and could enhance overall early lung cancer detection performance.
• Further optimization can lead to a standardized system for diverse thoracic CT dataset assessment.