A team of researchers from Massachusetts Institute of Technology (MIT) and Boston Children’s Hospital have developed one of the first-of-its-kind software methods for converting 3D MRI scans into tangible, handheld models to plan heart surgeries. Two major conclusions of this project are that:
- Routine use of 3D heart models for surgical planning is possible.
- Active learning approaches in which the user annotates uncertain regions of interest (ROI) can further reduce segmentation time.
Interactive patch-based segmentation algorithm
This achievement is unprecedented because it provides faster, more accurate results, and the scope of the project includes an interactive patch-based segmentation algorithm that incorporates user annotations.
The group used batch queries of uncertainty sampling to provide a simple strategy based on a uniform distribution of the input slices. They compared the two active learning methods (entire slices vs. smaller regions of interest) with several baseline approaches: uniform and random slice distribution. According to authors Danielle F. Pace, et al, they “implemented a strategy that exhaustively tries all possible new reference slices at each step and uses the gold-standard segmentation to add the slice that maximally reduces the segmentation error. This represents an iterative optimal greedy error reduction strategy with respect to maximizing improvement at each step.” The group validated its findings using four pediatric cardiac MRI images from patients with Double Outlet Right Ventricle (DORV).
The authors created an algorithm that utilizes information supplied by the user at each step of the segmentation. The user then manually segments a provided short-axis region while a patch-based method is used to update the segmentation volume. The algorithm reduces the number of interactions, yet directs the user to manually label part of the data deemed most informative instead of allowing the user to decide where to provide input. The group achieved amazing results when the expert segmented one-ninth of the total area of each cross-section.
Promising for future surgeries?
Every heart is different, so preparing for surgery can sometimes be difficult. Having a 3D model of the exact heart prepared for surgery could help surgeons deal with the specific issues.
According to an article by Larry Hardesty in MIT News, “The models could provide a more intuitive way for surgeons to assess and prepare the anatomical idiosyncrasies of individual patients.” The article also mentions that this fall, “seven cardiac surgeons at Boston Children’s Hospital will participate in a study intended to evaluate the models’ usefulness.”
Final Thoughts | 3D MRI Software Developments
Technology, especially in the health care industry, is increasing the possibilities for more accurate and positive outcomes. We look forward to seeing this 3D technology become more mainstream. In the meantime, if you offer MRI services at your facility, consider certifying your machine with RadSite accreditation. Accreditation demonstrates your organization is a trustworthy spot for consumers to receive best-in-class services and patient care.