Icon Legend

This session is not in your schedule.

This session is in your schedule. Click again to remove it.

Biomedical Imaging and Instrumentation

Biomedical Imaging and Instrumentation - Poster Session E

Poster R17 - Theranostic Pancreatic Ultrasound in Pediatric Clinical Medicine

Saturday, October 26, 2024

10:00 AM - 11:00 AM EST

Location: Exhibit Hall E, F & G

Presenting Author(s)

Gantt Meredith, BS, BA

Doctoral Researcher
Children's National Hospital
Washington, District of Columbia, United States

Primary Investigator(s)

KS

Karun Sharma

Head of Interventional Radiology
Children's National Hospital, United States

Co-Author(s)

JS

Joao Sanches

Professor
Tecnico Lisboa, United States

Introduction: Diagnostic imaging through CT has produced a plethora of disease diagnosis pipelines derived from imaging biomarkers relevant to specific disease pathologies. Ultrasound imaging, however, has limited research into its ability to produce reliable imaging biomarkers, despite its potential to serve as both a diagnostic tool and a clinical therapy for many diseases. One such disease, diabetes mellitus, can be diagnosed and detected via CT and MRI [1]. To our knowledge, no ultrasound-based imaging biomarkers exist for this pathology. Routine pancreatic imaging, especially of the pediatric pancreas, is not part of the ultrasound clinical standard, so this work first assessed the feasibility of incorporating pancreatic scans into current clinical abdominal protocols. Through an ultrasound imaging trial of healthy and diabetic pediatric patients, the pancreas was identified and analyzed with machine learning to begin identifying potential imaging biomarkers for diabetes mellitus. Ultrasound, in contrast to CT, can serve as both a diagnostic tool and a therapy (histotripsy, HIFU). Therapeutic ultrasound has been demonstrated to regulate insulin secretion in pancreatic beta cells [2]; therefore, safety and efficacy modeling was performed to ensure pediatric therapy is possible.

Materials and

Methods: B-mode contrast and doppler images of the pediatric pancreas were retrieved from an imaging trial at Children’s National Hospital in Washington, DC. Images were stratified for diabetic and healthy patients. From that point, anatomical features of the head, body, and tail were confirmed with a radiologist and recorded to derive a feasibility metric for imaging the pediatric pancreas in an abdominal ultrasound scanning protocol. Images were anonymized and features were extracted using three pre-trained deep learning models. Feature selection was performed, and these features will be fed into a support vector machine (SVM) to assess the feasibility of ultrasound images for diabetic detection. In conjunction with diabetic diagnosis, safety and efficacy multiphysics modeling was performed with an FDA-based acoustic modeling software in Matlab [3], validating appropriate targeting and non-harmful thermal effects to the region of the pediatric pancreas. Anatomical measurements and locations of the pancreas were derived from ultrasound scans of the abdomen, with specific locations of the pancreas offering the ability for personalized medicine and therapy.

Results, Conclusions, and Discussions: The head, body, and tail of the pancreas were visible in 100% of the CT scans evaluated for healthy and diabetic patients, with key morphological changes (texture serrations, atrophy, constriction) noted in diabetic images by interventional radiologists. With current abdominal ultrasound imaging, the pancreas was visible in approximately 80% of the available healthy and diabetic images, despite ultrasound sonographers reporting the pancreas was often a primary target of abdominal ultrasound scans. The lack of a standard scanning protocol for the pancreas within the ultrasound space can be filled with the information gathered from these two trials.

For feature extraction across modalities, both US and CT, the larger the number of input images, the larger the number of extracted features, even after refinement through PCA. While this observation associated with a larger sample size is a common feature of deep learning, finding the minimum number of images for a sonographer to acquire is an important finding. The minimum number of input images for pre-trained models like VGG16 and EfficientNetB0 was 4 for pancreatic US or CT images, regardless of the output node’s density. From this analysis, an informed review of the pancreas from an imaging perspective requires at least 4 images of either the head, body, and/or tail. To confirm the efficacy and safety of therapeutic ultrasound in the pediatric pancreatic space, transducer and sonication parameters previously shown to stimulate insulin release (800 kHz, 100% duty cycle, 0.5 W/cm^2), were modeled using a several centimeter focal point for acoustic stimulation. While observed heat was noticeable in surrounding organ groups (kidneys, spleen), the levels are below the clinical safety standard for pediatric care.

Acknowledgements (Optional): References
[1] Design and Rationale for the Use of Magnetic Resonance Imaging Biomarkers to Predict Diabetes After Acute Pancreatitis in the Diabetes Related to Acute Pancreatitis and Its Mechanisms Study From the Type 1 Diabetes in Acute Pancreatitis Consortium, Tirkes, T., et. al., July 2022.
[2] Effect of Therapeutic Ultrasound on the Release of Insulin, Glucagon, and Alpha‐Amylase from Ex Vivo Pancreatic Models, Castellanos, V., et. al., February 2021.
[3] High intensity focused ultrasound simulator, MATLAB Central File Exchange, Soneson, J., 2024.