A 3-D active shape model driven by fuzzy inference: application to cardiac CT and MR

Area of Science:

• Medical Imaging
• Computational Anatomy
• Cardiovascular Imaging

Background:

• Manual segmentation of cardiac imaging data (CT, MR) is time-consuming due to large volumes.
• Accurate quantification of left ventricular function requires precise segmentation.

Purpose of the Study:

• To develop and evaluate a 3-D active shape model (ASM) for semiautomatic cardiac left ventricle segmentation.
• To create a model adaptable to different imaging modalities (CT, MR) without retraining.

Main Methods:

• A 3-D active shape model (ASM) incorporating a fuzzy c-means based fuzzy inference system was developed.
• The model utilizes relative gray-level differences for region classification, enhancing cross-modality applicability.
• Evaluation involved 25 CT and 15 MR datasets, comparing automated contours to expert delineations.

Main Results:

• For CT, segmentation accuracy reached 82.4% (epicardium) and 74.1% (endocardium) within 5 mm error.
• For MR, accuracy was higher at 93.2% (epicardium) and 91.4% (endocardium).
• Volume analysis showed strong linear correlation (r(2) >/= 0.98) between manual and semiautomatic measurements.

Conclusions:

• The fuzzy inference 3-D ASM offers a robust and promising tool for semiautomatic cardiac left ventricle segmentation.
• The model's ability to segment across CT and MR without retraining makes it suitable for routine clinical use.