Database-guided breast tumor detection and segmentation in 2D ultrasound images

Combining CRF and Multi-hypothesis Detection for Accurate Lesion Segmentation in Breast Sonograms

“…Another competitor (DPM-Levelset) is a cascade of lesion detection and level set segmentation. Similar to [7], the result of detection is used as the bounding box of the lesion to initialize the shape model of level set. The maximal number of dynamic iterations is set to 500.…”

“…10% outliers are removed for all methods as in [7]. Let S be the segmented lesion region and G be the lesion in groundtruth.…”

“…The importance of each rule is learned and adjusted through training and practicing with a large number of samples. A notable work comes from Siemens researchers recently [7]. They train discriminative models of both texture and boundary, and combine them in the framework of Markov random field.…”

“…For ultrasound image, the problem is becoming more difficult because of the existence of similar tissues and artifacts, which makes the detection confidence less reliable. So it is risky to select one of the windows as the bounding box of the lesion [7] before investigating the interior.…”

HoliMAb: A holistic approach for Media–Adventitia border detection in intravascular ultrasound

“…Most of the proposed techniques are based on Conditional Random Fields (CRFs) or Markov Random Fields (MRFs). One example is in (Gupta et al, 2011), where CRFs is applied to the segmentation of fetal in ultrasound images; in (Schmidt et al, 2008) CRFs are applied to the detection of heart motion abnormality; in (Zhang et al, 2010) MRFs and Graph Cut are used in breast tumor detection. All these works can be compared to the approach presented in (Ciompi et al, 2011), where the tissue labeling relies on inference applied to a graphical model.…”