A dynamic lesion model for differentiation of malignant and benign pathologies

Area of Science:

• Medical Imaging
• Quantitative Pathology
• Computational Anatomy

Background:

• Malignant lesions exhibit greater invasiveness than benign ones, necessitating accurate differentiation.
• Quantifying lesion invasiveness is crucial for diagnosis and treatment planning.
• Existing methods for lesion characterization face limitations in capturing dynamic invasiveness.

Purpose of the Study:

• To propose a dynamic lesion model for quantifying invasiveness using image intensity changes.
• To develop a quantitative measure based on second-order derivatives (Hessian matrix eigenvalues).
• To evaluate the model's efficacy in differentiating malignant from benign lesions.

Main Methods:

• Utilized second-order derivatives (Hessian matrix) of image intensity at each voxel.
• Extracted quantitative measures (vector texture descriptors) from Hessian vectors derived from eigenvalues and eigenvectors.
• Applied the model to pathological datasets of colon polyps and lung nodules.

Main Results:

• The proposed dynamic lesion model effectively quantified lesion invasiveness.
• Vector texture descriptors successfully differentiated malignant from benign lesions in both datasets.
• The method significantly outperformed four state-of-the-art computational methods.
• The model's performance surpassed that of three expert radiologists.

Conclusions:

• The dynamic lesion model provides a robust quantitative measure for lesion invasiveness.
• This approach offers a promising tool for improving the accuracy of differentiating malignant from benign lesions.
• The findings suggest potential applications in radiological diagnosis and image-guided interventions.