Technology

Contouring, or segmentation, of medical images from different modalities, e.g. MRI, CT or Ultrasound, is a major activity in disease diagnosis, treatment planning and therapy.

Currently, clinicians must rely on laborious manual segmentation for results where either there is no software or where existing software provide unsatisfactory results. The inadequacies of current automated solutions also mean that valuable volumetric information cannot be extracted: utilization of this information could vastly improve the clinician’s ability to plan and track patient therapy. The Segasist platform provides a new approach to auto-contouring to break through existing deficiencies of conventional segmentation technologies. Through calibration via “training” with gold standard images, Segasist makes contouring decisions to provide high-quality results. In other words, the results of the software get better the better it is trained or the more it is used. The software observes, captures and saves contouring and editing preferences over time, and applies this knowledge to contour new images. Segasist learns to contour like an expert would.

Segasist surpasses any existing upper bound on contouring accuracy by learning to optimally guide the segmentation process using gold-standard images in order to reach higher agreement with the expert user.

Segasist also captures multiple User Profiles - each storing distinct contouring preferences - and uses the accumulated knowledge from all of these profiles to create a "Best-Practices" rule set. This unique characteristic of Segasist has two major advantages:

• Segasist can provide consensus contours to reduce differences among experts (decreasing inter-observer variability).Note: Since Segasist keeps User Profiles, it can generate consensus contours at any time for any image even in absence of all users.

• Segasist can warn the user about inaccurate contours and non-standard edits to encourage adherence to best practices.

Figure 1. Learning through calibration with gold standard images: The software generates more accurate contours the more images are processed (red curve). In addition, if the information of the same patient is used (volume data of the same patient being segmented), the accuracy can be maintained at a very high level (blue curve).

» Watch a video showing the learning effect

OMISA: Beyond the Modality Barrier

Segasist software uses a combination of unique, proprietary algorithms referred to collectively as the Omni-Modality Intelligent Segmentation Agent (OMISA). This software agent is the core of Segasist Learning Engine with several advantages over commercially available methods:

• It is the only segmentation method that can incorporate real-time, continuous calibration.
• It requires far less manual segmentation to achieve high accuracy.
• It can drastically reduce the time required to achieve volumetric analysis such as precise tracking or tracing tumor/organ volumes.
• It is the only software that remembers clinician preferences and interpretation.
• It can help reduce inter-observer variability.
• It is operable on multiple modalities and clinical cases.

For more information about Segasist's accuracy and learning capabilities, please view our White Papers.