Accelerating Chan-Vese model with cross-modality guided contrast enhancement for liver segmentation

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Processing

Background:

• Accurate liver segmentation is crucial for clinical diagnosis but is hindered by noise and low contrast in CT scans.
• Existing segmentation algorithms, like Chan-Vese, are robust to noise but computationally slow, limiting real-time applications.
• Low contrast in CT liver images further degrades segmentation quality.

Purpose of the Study:

• To develop an efficient and accurate liver segmentation method for computed tomography (CT) scans.
• To address the speed limitations of the Chan-Vese algorithm through GPU parallelization.
• To improve segmentation quality by implementing a cross-modality guided contrast enhancement pre-processing step.

Main Methods:

• Implemented a GPU-accelerated version of the Chan-Vese active contour model for liver segmentation.
• Integrated a cross-modality guided contrast enhancement technique as a pre-processing step.
• Evaluated segmentation performance using metrics such as Dice coefficient, sensitivity, and accuracy on CT abdominal scans.

Main Results:

• The GPU implementation of Chan-Vese achieved an average speedup of 99.8 times (without enhancement) and 14.6 times (with enhancement) compared to CPU.
• Liver segmentation accuracy improved significantly with contrast enhancement, with Dice scores rising from 0.656 to 0.877.
• Enhanced images showed improved sensitivity (0.816 to 0.964) and accuracy (0.822 to 0.956).

Conclusions:

• GPU acceleration and contrast enhancement dramatically improve the speed and accuracy of CT liver segmentation.
• The proposed method offers a viable solution for real-time, high-quality liver segmentation in clinical settings.
• This approach enhances the clinical utility of CT imaging for liver-related conditions.