Computer-aided diagnosis of soft-tissue tumors using sonographic morphologic and texture features

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Soft-tissue tumors require accurate differentiation between benign and malignant types for appropriate patient management.
• Sonography is a key imaging modality for soft-tissue tumor assessment, but distinguishing between benign and malignant lesions can be challenging.

Purpose of the Study:

• To develop and evaluate a computer-aided diagnosis (CAD) system for assessing sonographic morphologic and texture features of soft-tissue tumors.
• To compare the diagnostic performance of the CAD system with that of experienced radiologists.

Main Methods:

• A retrospective analysis of 114 soft-tissue tumors (73 benign, 41 malignant) was performed.
• Morphologic and gray-level co-occurrence matrix texture features were extracted from sonographic images.
• Linear discriminant analysis (LDA) and a multilayer neural network (MLP) were employed as classifiers.

Main Results:

• The CAD system, combining morphologic and texture features, achieved high diagnostic performance.
• LDA classifier yielded an accuracy of 89.5% and an A(z) value of 0.96.
• MLP classifier achieved an accuracy of 88.6% and an A(z) value of 0.95, both outperforming the average radiologist's performance (A(z) 0.74-0.86).

Conclusions:

• The developed CAD system effectively distinguishes between benign and malignant soft-tissue tumors using sonographic features.
• The CAD system can serve as a valuable second opinion tool, potentially aiding in diagnosis and reducing unnecessary biopsies.