Adaptive Estimation of Active Contour Parameters Using Convolutional Neural Networks and Texture Analysis.

Area of Science:

• Medical imaging
• Image segmentation
• Machine learning in radiology

Background:

• Accurate segmentation of liver lesions is crucial for diagnosis and treatment planning.
• Existing active contour and CNN-based methods have limitations in handling low contrast, heterogeneous, and noisy lesion images.

Purpose of the Study:

• To develop a generalized and fully automatic level set segmentation approach for liver lesions.
• To introduce a novel method for adaptive estimation of active contour parameters integrated with CNNs.

Main Methods:

• A convolutional neural network (CNN) estimates the initial contour position relative to the lesion.
• CNN output probabilities adaptively guide active contour parameter calculation.
• An iterative process refines contour window size based on lesion characteristics.

Main Results:

• The proposed method significantly outperformed state-of-the-art CNN-based and active contour techniques.
• An average Dice similarity coefficient improvement of 0.27 was achieved across a dataset of 164 MRI and 112 CT liver lesion images.
• Significant improvements (Dice 0.24, p < 0.001) were observed on challenging lesion subsets.

Conclusions:

• The novel adaptive active contour segmentation method offers superior performance for liver lesion segmentation.
• This approach effectively addresses challenges posed by low contrast, heterogeneity, and noise in medical images.
• The integration of CNNs with adaptive parameter estimation represents a significant advancement in automated medical image segmentation.