THE JOURNAL OF KMUTNB

Volume 32, No. 04, Month OCTOBER, Year 2022, Pages 865 - 878

Boundary extraction of multichannel magnetic resonant images based on a model of particle motion in a vector field derived from local color distance images

Wutthichai Phornphatcharaphong, Nawapak Eua-Anant

Abstract Download PDF

Each type of magnetic resonance (MR) image has distinctive features suitable for the diagnosis of different lesions, in which doctors can analyze each MR image separately or combine several MR images, of different types, to gain more information. In the latter case, the combination of MR images of different types results in a multichannel MR image. Analyzing a multichannel MR image is far more difficult than analyzing a single MR image due to the larger data volume and the complexity of inter-channel data relation. This article presents one of the methods for analyzing a multichannel MR image, namely, boundary extraction of a multichannel MR image using a model of particle motion in a vector field derived from local color distance images (PMLCD). The proposed method is developed from the boundary extraction algorithm based on particle motion in a vector image field (PMVIF), which could previously only be applied to grayscale images. This paper proposes two new vector fields, used in the model, which can be computed directly from a multichannel MR image: 1) the normal compressive vector field derived from the center-to-centroid vectors of local color distance images. 2) The edge vector field derived from the Hamiltonian gradient-like vector field modified from the normal compressive vector field. Performance evaluation results of multichannel MR image segmentation using Rand Index (RI), Global Consistency Error (GCE), Normalized Variation of Information (NVI), Boundary Displacement Error (BDE), and Dice Coefficients indicate that the proposed method has good performance and fast computation time.

Keywords

Boundary Extraction; Particle Motion; Local Color Distance Images; Magnetic Resonant Images; Multichannel Images

---