Hessian-MRLoG: Hessian information and multi-scale reverse LoG filter for pulmonary nodule detection

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Oncology

Background:

• Computer-aided detection (CADe) is crucial for early lung cancer diagnosis using computed tomography (CT).
• Existing CADe methods struggle with low contrast, blood vessel interference, and classification issues, leading to low detection rates and high false-positive rates (FPR).

Purpose of the Study:

• To develop a novel Hessian-MRLoG method for improved pulmonary nodule detection.
• To address the limitations of existing CADe techniques in lung CT image analysis.

Main Methods:

• A novel Hessian-MRLoG method combining Hessian information and multi-scale reverse Laplacian of Gaussian (LoG) filtering was developed.
• A multi-scale reverse Laplacian of Gaussian (MRLoG) was constructed to address intensity and shape inconsistencies.
• An adjustment factor was introduced to optimize second-order partial derivatives, and a novel elliptic filter was designed.

Main Results:

• The Hessian-MRLoG method achieved 93.6% accuracy on the LUNA16 dataset.
• The method produced a low false-positive rate of 1.0 false positives per scan (FPs/scan).
• Experimental results demonstrate improved detection rates and significantly reduced FPR.

Conclusions:

• The proposed Hessian-MRLoG filtering method is effective for pulmonary nodule detection.
• This technique shows potential for application in detecting, localizing, and labeling other lesion areas in medical imaging.