1/16/2025 Ben Libman
Ultrasound Computed Tomography (USCT) offers a promising alternative to traditional mammography, providing non-invasive imaging without radiation. A novel method called image-to-image learned reconstruction (IILR) enhances USCT imaging by combining deep learning with computationally efficient modalities like traveltime tomography and reflection tomography. By making advanced imaging more accessible and efficient, IILR has the potential to improve early breast cancer detection and outcomes across diverse clinical settings.
Written by Ben Libman
Breast cancer is a devastating disease. A critical step toward a positive outcome is early detection, which significantly improves a patient’s prognosis. This underscores the importance of imaging. The method doctors use to detect tumors must be accurate and accessible to all. The most common screenings use mammography or tomosynthesis, but these methods are prone to false positives, can be uncomfortable, and even expose patients to low doses of ionizing radiation.
A potential alternative is Ultrasound Computed Tomography (USCT). USCT is an emerging, non-invasive imaging modality which offers several advantages over traditional mammography: it is less uncomfortable, eliminates radiation exposure, and provides fewer false positives, especially for individuals with particularly dense breast tissue. USCT uses sound waves to generate three-dimensional images that depict the acoustic properties of breast tissue, such as the speed-of-sound (SOS) distribution. These SOS “maps” enable researchers to differentiate between normal, benign, and cancerous tissues. As it is an emerging field, there are multiple ways to take the information generated by USCT and create an image of the tissue. One technique used to create accurate, high-resolution SOS maps is known as full-waveform inversion (FWI). FWI is highly accurate; however, it is computationally intensive and time-consuming. This makes FWI impractical in many clinical settings, especially in low-resource environments.
A new method for forming USCT images may be able to close that accessibility gap. The research, reported by lead author Gangwon Jeong, was recently published by the Institute of Electrical and Electronics Engineers (IEEE). He worked with his advisors, Donald Biggar Willett Professor in Engineering and Bioengineering Department Head Mark Anastasio and Dr. Umberto Villa from UT Austin, along with their collaborators Dr. Nebojsa Duric, Trevor M. Mitcham, and Fu Li. This novel technique, called image-to-image learned reconstruction (IILR), uses deep learning technologies to reconstruct high-resolution SOS maps. Specifically, it combines images produced by computationally efficient USCT imaging modalities—traveltime tomography (TT) and reflection tomography (RT).
TT provides a quantitative SOS distribution of tissue but with low resolution, which limits its use for tumor detection. RT provides high-resolution details of tissue boundaries but does not offer quantitative SOS information. By combining the images produced by these two complementary modalities as inputs to a deep neural network an accurate SOS map is generated with resolution comparable to that generated by FWI, with only a fraction of the computational cost. IILR even showed some improvements- it was less prone to errors in the image, known as “artifacts.” Critically, this model is significantly faster, and can produce images much more quickly than FWI. Coupled with its reduced computing power requirements, IILR becomes a potentially vital tool for accessibility in clinics with fewer resources.
IILR shows significant promise, but there is still progress to be made. The traditional FWI method is slightly better at detecting tumors, although IILR showed significant improvement when the training process was adjusted to focus more on tumor areas.
Though challenges remain, the potential of IILR is exciting. This technology could bring faster results more affordably to places with fewer resources and do so with fewer errors. That would mean cancer being detected sooner and more often, treatments starting earlier, and better health outcomes for everyone. “Everyone should have access to the best treatment, regardless of their geography or socioeconomic status,” said professor Anastasio. “IILR represents the next step in that vision.”
Read the full paper here.
Mark Anastasio is the Donald Biggar Willett Professor in Engineering and Department Head of Bioengineering, with affiliate appointments at the Beckman Institute, Carle-Illinois College of Medicine, the Department of Computer Science, and the Department of Electrical and Computer Engineering.
Nebojsa Duric is a Professor in the Department of Imaging Sciences at the University of Rochester Medical Center, with joint appointments in the Department of Biomedical Engineering and Department of Imaging Sciences (Vice Chair for Research.)
Umberto Villa is a Research Associate Professor in the Oden Institute for Computational Engineering and Sciences at the University of Texas at Austin.
Trevor M. Mitcham is a Postdoctoral Research Associate in the Department of Imaging Sciences at University of Rochester Medical Center.
Gangwon Jeong and Fu Li are Ph.D. candidates in the Department of Bioengineering at the University of Illinois at Urbana-Champaign.