Automatic initialization of active contours and level set method in ultrasound images of breast abnormalities. (July 2018)

• Record Type: Journal Article
• Authors: Rodtook, Annupan; Kirimasthong, Khwunta; Lohitvisate, Wanrudee; Makhanov, Stanislav S.
• Is Part Of: Pattern recognition. Volume 79(2018:Jul.)
• Page Start: 172
• Page End: 182
• Publication Date: 2018-07
• Subjects: Ultrasound image segmentation; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2018.01.032
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: The paper presents an original method to initialize the active contours and the level set method in ultrasound images of breast cancer using walking particles. The method tested against five state-of-the-art initialization methods shows the advantage in terms of the accuracy or/and processing time. The combination of the proposed walking particles and the level set method is the most suitable for segmentation of breast tumors. Abstract: We propose a novel initialization method designed for active contours (AC) and the level set method (LSM), based on walking particles. The algorithm defines the seeds at converging and diverging configurations of the corresponding vector field. Next, the seeds "explode", generating a set of walking particles designed to differentiate between the seeds located inside and outside the object. The exploding seeds method (ESM) has been tested against five state-of-the-art initialization methods on 180 ultrasound images from a database collected by Thammasat University Hospital of Thailand. The set of images was additionally partitioned into malignant tumors, fibroadenomas and cysts. The method has been tested for each of those cases using the ground truth hand-drawn by leading radiologists of the hospital. The competing methods were: the trial snake (TS), centers of divergence (CoD), force field segmentation (FFS), Poisson Inverse Gradient Vector Flow (PIG), and quasi-automated initialization (QAI). The numerical tests demonstrated that CoD and FFS failed on the selected test images, whereas the average accuracy of PIG and QAI were lower than that achieved by the proposed method for both AC and the LSM. The LSM combined with the ESM provides the best results.