Skip to main content
2024 BMES Annual Meeting Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.
Back

Biomedical Imaging and Instrumentation

Biomedical Imaging and Instrumentation - Poster Session C

Poster G2 - Low-cost layered ultrasound phantoms to simulate neonatal muscle and fat

Friday, October 25, 2024
10:00 AM – 11:00 AM EST
Location: Exhibit Hall E, F & G

Introduction:  Ultrasound phantoms have a variety of applications in medical image analysis, including for calibrating medical equipment under controlled environments, and for practicing ultrasound-guided procedures. Research and development related to muscle and fat phantoms for ultrasound imaging has primarily focused on adult applications. However, there is growing interest in adapting and optimizing these phantoms for newborn imaging. Particularly, the difficulties of clinical recruitment and routine measurements for younger populations have contributed to the shortage of clinical datasets for this population. Increased access to ultrasound datasets, both clinical and phantom, have applications in neonatal intensive care units (NICUs) to aid in early diagnosis and monitoring of conditions such as neonatal adiposity and muscle development. Specifically, we are interested in using neonatal ultrasound datasets to train segmentation models to predict neonatal body composition based on tissue thickness and cross-sectional area. However, training a model requires large and diverse datasets with varying muscle and adipose thickness. We propose an alternative to using clinical data collection for our model training purposes: neonatal phantoms. In this paper, we present a procedure for constructing affordable phantoms that mimic newborn anatomy for ultrasound imaging that is comparable to clinical data. We envision this procedure to be utilized to create large quantities of neonatal phantom datasets that can provide an alternative data source for medical image analysis.

Materials and

Methods:  The following methodology was used to develop diverse ultrasound phantoms for newborns. Three types of phantoms were constructed to represent different scanning locations on newborns: biceps (brachii and brachialis), quadriceps (rectus femoris and vastus intermedius), and abdomen (rectus abdominis). Tissue-mimicking materials (TMM) were constructed separately and layered to mimic the newborn anatomy. Muscle TMM was made with 8% agarose and 2% aluminum oxide by weight in boiling distilled water. The TMM was fully hardened at 0˚C for 10 minutes before being cut. Fat TMM was made with styrene ethylene butylene styrene (SEBS) and 1% psyllium husk by weight. SEBS was heated to 200˚C for 30 minutes and the psyllium husk was added. Bone TMM was made with plaster (calcium sulfate hemihydrate, calcium carbonate, and crystalline silicon) with dimensions proportional to newborn skeletal anatomy. Aquasonic Clear ultrasound gel was layered in between each TMM layer to prevent air bubbles. The specific composition for each TMM was determined by creating phantoms with agarose varying from 6-8%, aluminum oxide varying from 1-10%, and psyllium husk varying from 0.5-5%. Ultrasound images of the phantoms were collected using a portable Clarius L20 HD3 scanning system and were qualitatively compared to clinical data that represented comparable anatomy. We determined the optimal concentrations for each TMM by qualitatively comparing the layered phantom ultrasound quality compared to clinical datasets.

Results, Conclusions, and Discussions: Example ultrasound phantom and clinical images for biceps, quadriceps, and abdomen are shown in Figure 1. When comparing each phantom and clinical image, we note similar anatomical structures of the skin, adipose tissue, muscle, and bone. In particular, the abdominal phantom images show similar echogenicity in both muscle and fat layers, as compared to the respective clinical image. In addition, the phantom images of the biceps and quadriceps show notable morphological similarities for layers of both muscle and fat. For the purposes of creating large ultrasound datasets of the neonatal body parts to train models to predict body composition, these results provide evidence that phantom datasets may be an alternative data source. Especially if the medical image analysis involves the segmentation of tissue layers, we predict that phantom datasets will be a cheaper, quicker, and more efficient way to train models. Additionally, we are able to control the thickness of each layer of TMM, which provides an advantage when acquiring diverse ultrasound datasets. By customizing the TMM thicknesses, positioning, and orientation of each layer, we are able to customize our datasets easily. However, when evaluating our neonatal phantom images for purposes outside of tissue segmentation, our phantoms could be improved by adding intramuscular fat striations within the layers of muscle TMM. Moreover, the blurriness of the ultrasound images can be improved by using a higher-quality ultrasound imaging modality.

In summary, our low-cost construction method allows us to create a large number of neonatal phantoms that can mimic the echogenicity of commonly imaged tissues in the newborn population. This can address the scarcity of ultrasound data for this population for many applications, one being for training segmentation models to predict body composition. In the future, we may qualitatively assess our phantom construction performance by using our neonatal phantom datasets to train segmentation models and compare their performance against models training with neonatal clinical datasets.

Acknowledgements (Optional):