Sri-Rajasekhar Kothapalli (left), assistant professor of biomedical engineering, and Sumit Agrawal (right), a graduate student in Kothapalli’s lab at the time, led the development of a platform that combines ultrasound and photoacoustic imaging to simulate human deep tissue. Credit:Penn State
Combined ultrasound-photoacoustic simulations help AI improve image quality
5/05/2022
By Ashley J. WennersHerron
UNIVERSITY PARK, Pa. — Combined ultrasound and photoacoustic (USPA) imaging can provide structural, functional and molecular information of deep biological tissue in real time, but its quantitative performance is affected by several unknown tissue parameters. To address this limitation, Penn State researchers established a simulation platform capable of modeling USPA imaging device performance for a given clinical application to generate large-scale training data sets for artificial intelligence to improve image quality.
Led by Sri-Rajasekhar Kothapalli, assistant professor of biomedical engineering, the researchers published their approach in Photoacoustics.
While conventional ultrasound imaging provides anatomical information, photoacoustic imaging involves exciting light-absorbing tissue chromophores, such as hemoglobin and melanin, to expand and generate photoacoustic waves. The photoacoustic waves propagate and can be detected by the same ultrasound transducer that is used for conventional ultrasound imaging to map 3D molecular information of the internal tissue.
“Over the past decade, dual-modality USPA imaging has been translated to several clinical applications and received FDA approval for breast cancer screening,” Kothapalli said. “However, USPA imaging commonly suffers from depth and wavelength-dependent optical and acoustic attenuation, which affects the visibility of deep tissue targets. Imaging performance can also be limited by unknown optical and acoustic differences, as well as reflection artifacts. Novel instrumentation, image reconstruction and AI methods are currently being investigated to overcome these limitations and improve the USPA image quality. Effective implementation of these approaches requires a reliable USPA simulation tool capable of generating both ultrasound-based anatomical and photoacoustic-based molecular contrasts of deep biological tissue.”
The researchers developed such a tool for co-simulating ultrasound and photoacoustic images, by integrating elements from two open-source software resources: k-Wave and NIRFAST. The former provides ultrasound simulations, while the latter provides finite element models of light propagation inside a tissue scattering medium. Together, with modifications from Kothapalli and his team to account for USPA-specific parameters such as aperture size and frequency of detector arrays, the resulting platform can simulate deep tissue imaging of human organs.
“Besides combining ultrasound and photoacoustic imaging, our simulation platform enables design of realistic digital tissue phantoms, such as the human prostate and breast, which can be encoded with heterogenous optical and acoustic properties of different biological structures,” said first author Sumit Agrawal, who contributed to the research as a graduate student in Kothapalli’s lab. He earned his doctorate in 2021 and is now an ultrasound imaging systems engineer with Exo. “Our results demonstrated that simulated ultrasound and photoacoustic images of the human prostate and the human finger closely matched with respective experimental images of human volunteers. We further expanded the simulation platform to generate application-specific, large-scale ultrasound and photoacoustic datasets to train AI algorithms to overcome the challenges of USPA imaging.”
In one application, the researchers trained an AI algorithm to detect and remove specific reflection artifacts in photoacoustic images of the human finger. In another application, the researchers improved the quantitative accuracy of deep tissue oxygen saturation. The simulation platform is expected to open the door for developing and integrating reliable AI algorithms to develop smart USPA devices for several biomedical applications, according to Kothapalli.
The complete USPA simulation codes and user guides are open source and available here.
Other authors include Thaarakh Suresh and Ajay Dangi, Department of Biomedical Engineering, and third-year undergraduate student Ankit Garikipati, Department of Electrical Engineering. Kothapalli is also affiliated with the Penn State Cancer Institute, and Suresh is also affiliated with the University of Pittsburgh’s Department of Bioengineering.
The National Institutes of Health, the Penn State Cancer Institute, the Leighton Riess Graduate Fellowship and the NVIDIA Corporation supported this work.
Share this story:
