Edit this page

NA-MIC Project Weeks

Fetal Ultrasound Simulation for Delivery Training

Key Investigators

Clinical collaboration

Medical Background

Intrapartum assessment of the fetal head position and pelvic station are essential in the management of labor. Precise knowledge of these parameters will assist in the correct identification of normal vs. abnormal labor patterns, and among the latter indicate when medical or operative intervention may be required.

Identifying the fetal position is specially important during forceps-assisted delivery. Forceps are slick metal objects that resemble big spoons or tongs. They are contoured to fit the baby’s head and are placed around the infant’s head. Depending on the position of the fetus’s head, the techniques and manouvers used by obstetricians are different.

Ultrasound imaging is commonly used to estimate the fetal position. In a first step, the clinician identifies the fetus spine using ultrasound to infer the position of the fetus body. Then, the occiput is localized to determine the exact position of the head.

drawing

drawing

Project Description

The goal of this project is to generate fake ultrasound images that can be used during medical training. These images should be as realistic as possible to enable trainees to correctly identify the fetus position in the simulated ultrasound images.

PLUS Toolkit already offers a tool to generate fake ultrasound images based on 3D models. However, a general problem is that one needs to provide acoustic properties of each 3D model/tissue for the ultrasound image generation.

Objective

  1. Test ultrasound simulation tool offered by PLUS Toolkit and visualize simulated images and models in 3D Slicer.
  2. Create a tool that automatically assigns acoustic properties to each 3D model and generates the according config file for the PLUS ultrasound image generation server.
  3. Calculate acoustic properties for fetal ultrasound simulation and improve realism of simulated ultrasound images.
  4. Discuss ultrasound simulation technology with the 3D Slicer community. Other methods? AI?

Approach and Plan

Development of tool to assign tissue acoustic properties:

  1. Use the Total Segmentator to create the 3D models with assigned organ definition based on CT data.
  2. Create and define a lookup table (LUT) that assigns the accoustic properties to each organ label.
  3. Develop a tool or module that generates the config file based on the models and the LUT.

Progress and Next Steps

  1. Created a LUT with accoustic properties (density, speed of sound, accoustic attenuation) of 112 different tissue types.
  2. Created a simple Matlab script to generate an XML config file for PLUS with the according accoustic properties for different models based on Total Segmentator

drawing

  1. Generate scene with virtual models of mother and fetus.

drawing drawing

  1. NEXT STEP: Test ultrasound image simulation assigning acoustic properties to tissues.

Conclusion & Outlook

A central challenge regarding surface model based ultrasound simulation, i.e., the generation of realistic mesh models of internal anatomy, is greatly alleviated by the utilization of the Total Segmentator module that allows for the automated segmentation of multiple tissues and organs with appreciabe accuracy. In terms of various sound propagation and ray-tracing algorithms used ultrasound simulations, the classification of the respective segmentations enables the direct assignment of acoustic tissue properties that were researched and gathered from the related literature to build an tissue-specific acoustic lookup table. All of this is in stark contrast to previously proposed approaches that employed laboriously hand-crafted mesh models and manual fine-tuning of acoustic parameters.

Still, a major subject for future research remains: Usage of tissue segmentations delineating the outer border of structures within the framework of surface-based ultrasound simulation, e.g. as implemented in PLUS, disregards intra-structural heterogeneity of tissues and organs. Depending on the respective applications, this may lead to undesirable low simulation-fidelity. Hence, we intend to exploit image intensities within source images of the segmentations to perform intensity-based modifications of the acoustic parameters assigned within segmentations.

References