Multifeature landmark-free active appearance models: application to prostate MRI segmentation.

Area of Science:

• Medical image analysis
• Computational anatomy
• Machine learning for medical imaging

Background:

• Active shape models (ASMs) and active appearance models (AAMs) are established methods for medical image segmentation.
• These models typically rely on image-derived landmarks, necessitating accurate triangulation and alignment.
• Existing AAMs have limited exploration of multiple image-derived attributes (IDAs) for enhanced segmentation.

Purpose of the Study:

• To present a novel AAM methodology using levelset representation to overcome landmark-related issues.
• To incorporate multiple IDAs into the AAM framework for improved segmentation accuracy.
• To evaluate the proposed multifeature landmark-free AAM (MFLAAM) for prostate segmentation in T2-weighted MRI volumes.

Main Methods:

• Developed a novel AAM using levelset representation for landmark-free segmentation.
• Integrated multiple image-derived attributes (IDAs) into the AAM framework.
• Employed an efficient algorithm to identify optimal IDAs for segmentation accuracy.
• Applied the MFLAAM to prostate segmentation from T2-weighted MRI volumes.

Main Results:

• The levelset MFLAAM achieved a mean Dice accuracy of 88% ± 5% and a mean surface error of 1.5 mm ± 0.8 mm.
• Segmentation time was approximately 150 seconds per volume.
• The MFLAAM demonstrated statistically significant improvements over a state-of-the-art AAM (Dice 86% ± 9%, surface error 1.6 mm ± 1.0 mm).
• Results were superior to several recent state-of-the-art prostate MRI segmentation methods.

Conclusions:

• The proposed levelset-based MFLAAM offers an accurate and efficient approach for prostate segmentation.
• The landmark-free methodology and incorporation of multiple IDAs significantly enhance segmentation performance.
• This novel framework holds promise for advancing medical image segmentation in clinical practice.