Novel morphometric based classification via diffeomorphic based shape representation using manifold learning.

Area of Science:

• Medical Imaging Analysis
• Computational Anatomy
• Biomedical Informatics

Background:

• Anatomical morphology offers crucial diagnostic insights for diseases.
• Existing implicit morphological features are often domain-specific and may miss subtle shape variations.
• Computerized decision support systems benefit from advanced feature extraction for improved diagnostic accuracy.

Purpose of the Study:

• To present a novel framework for extracting Diffeomorphic Based Similarity (DBS) features.
• To capture subtle morphometric differences in anatomical structures beyond traditional implicit features.
• To evaluate the efficacy of DBS features in clinical disease classification tasks.

Main Methods:

• Representing object morphology using the medial axis model.
• Employing a cluster-based diffeomorphic registration scheme for medial axis model comparison.
• Utilizing manifold learning (Graph Embedding) to identify nonlinear dependencies and calculate DBS.
• Training a support vector machine (SVM) classifier with DBS features.

Main Results:

• DBS features demonstrated superior classification of shapes within the same class compared to implicit features.
• Achieved 83 +/- 4.47% accuracy in distinguishing benign from malignant breast lesions on DCE-MRI.
• Exceeded 80% accuracy in differentiating intermediate Gleason grades of prostate cancer from gland morphology.

Conclusions:

• The proposed DBS framework effectively captures subtle morphometric variations crucial for disease diagnosis.
• DBS features enhance the performance of machine learning classifiers in medical image analysis.
• This approach holds promise for improving diagnostic accuracy in complex clinical scenarios like prostate cancer grading and breast lesion classification.