A novel vessel enhancement method based on Hessian matrix eigenvalues using multilayer perceptron

Vessel segmentation is a critical aspect of medical image processing, often involving vessel enhancement as a preprocessing step. Existing vessel enhancement methods based on eigenvalues of Hessian matrix face challenges such as inconsistent parameter settings and suboptimal enhancement effects across different datasets.

This paper aims to introduce a novel vessel enhancement algorithm that overcomes the limitations of traditional methods by leveraging a multilayer perceptron to fit a vessel enhancement filter function using eigenvalues of Hessian matrix. The primary goal is to simplify parameter tuning while enhancing the effectiveness and generalizability of vessel enhancement.

The proposed algorithm utilizes eigenvalues of Hessian matrix as input for training the multilayer perceptron-based vessel enhancement filter function. The diameter of the largest blood vessel in the dataset is the only parameter to be set.

Experiments were conducted on public datasets such as DRIVE, STARE, and IRCAD. Additionally, optimal parameter acquisition methods for traditional Frangi and Jerman filters are introduced and quantitatively compared with the novel approach. Performance metrics such as AUROC, AUPRC, and DSC show that the proposed algorithm outperforms traditional filters in enhancing vessel features.

The findings of this study highlight the superiority of the proposed vessel enhancement algorithm in comparison to traditional methods. By simplifying parameter settings, improving enhancement effects, and showcasing superior performance metrics, the algorithm offers a promising solution for enhancing vessel parts in medical image analysis applications.