Cancer

Glioblastoma multiforme (GBM) is a highly aggressive brain cancer that often recurs within months of treatment, which typically involves surgery, radiation therapy (RT), and chemotherapy. We are using a powerful MR Spectroscopic Imaging technology known as SPICE (SPectroscopic Imaging by exploiting spatiospectral CorrElation) to help guide the delineation of a biological target volume to improve image-guided RT. SPICE effectively integrates rapid scanning, sparse sampling, quantum simulation of molecule resonance structures, and machine learning to enable rapid high-resolution MRSI.

Investigator: Chao Ma

Imaging, primarily mammography, is widely used for breast cancer screening and is essential for early detection, accurate diagnosis, and effective treatment planning. Imaging techniques enable detailed visualization of breast tissue, helping to identify tumors, assess their size and spread, and guide biopsies and surgical interventions.

Digital Breast Tomosynthesis (DBT) provides higher resolution and 3D-like breast images, but the large data volume complicates accurate cancer detection. To address this, we developed SIFT-DBT, a method that uses advanced techniques to identify abnormal images by focusing on smaller patches while maintaining high detail.

Investigator: Nicha Dvornek

Breast density—how much fibroglandular tissue there is compared to fat—matters for two reasons: it raises the risk of breast cancer and makes mammograms harder to read. In women with mostly fatty breasts, mammograms find 85–92% of cancers. In very dense breasts, that drops to 30–68%, and the cancer risk is 4–6 times higher. Radiologists use a scale called BI-RADS® to classify density, but because it relies on individual judgment, results can vary.

To help, our team worked with Visage Imaging, Inc. to create an AI tool that predicts what a group of radiologists would likely agree on. The result appears directly in the Visage software doctors already use to read mammograms. In less than two years, the tool was approved in the U.S., Europe, Australia, and Canada and has been in everyday use at Yale New Haven Health since 2021. Visage Breast Density is now a practical solution helping patients every day.

Lewin J, et al., PACS-integrated machine learning breast density classifier: clinical validation. Clin Imaging. 2023 Sep;101:200-205. doi: 10.1016/j.clinimag.2023.06.023.

Investigators: MingDe Lin, Liane Philpotts, John Lewin

We are developing molecular MRI probes to visualize the immuno-metabolic crosstalk in liver tumors. By integrating locoregional delivery of targeted contrast agents with imaging biomarkers of immune cell infiltration and metabolic activity, we enable noninvasive, longitudinal monitoring of the tumor microenvironment. These imaging strategies are expected to transform liver cancer management by providing early indicators of treatment response and uncovering key mechanisms of immunotherapy resistance, paving the way for precision-guided interventions.

Find more information at the Interventional Oncology Research Lab website.

Investigator: Julius Chapiro

Imaging is crucial to guiding the treatment of cancer. In radiation therapy (RT), CT is critical for developing external beam therapy plans, as well are assisting with patient setup. More recently, PET and MR have contributed to planning and setup also.

We are developing an automated method using deep neural networks (DNN) for contouring tumors, crucial for RT planning, particularly in sarcomas and head-and-neck tumors. This method aims to predict gross tumor volume (GTV) and clinical target volume (CTV) from multi-modality images while accounting for observer variability. By using contours from multiple readers, the DNNs will generate confidence maps to (1) provide accurate initial contours for refinement, improving efficiency, and (2) highlight challenging areas needing reader attention. We are evaluating conventional U-Net, diffusion networks, and transformers for this task.

Investigators: Georges El Fakhri, Thibault Marin, Chao Ma

Prostate cancer is a leading malignancy in men, with imaging playing a crucial role in diagnosis, risk stratification, and treatment planning. Researchers at Yale BioImaging are developing AI-powered image analysis software and advanced MRI devices to enhance diagnostic accuracy, treatment precision, and patient outcomes in collaboration with Yale School of Medicine's multidisciplinary prostate cancer team.

Investigators: John Onofrey, Thibault Marin

We are also embarking on a project to improve the dosing for ¹⁷⁷Lu-PSMA, a radiopharmaceutical therapy (RPT) compound used in the treatment of metastatic castration-resistant prostate cancers (mCRPC). See more in the “Theranostics” section of this page.

Investigators: John Onofrey, Thibault Marin

Theranostics merges diagnostic imaging with targeted therapy to enhance and personalize medical treatment. Molecules targeting disease sites can be labeled with photon-emitting isotopes for imaging and beta- or alpha-emitting isotopes for therapy, improving treatment efficacy by accurately targeting diseases and minimizing side effects. In prostate cancer, PSMA-targeting molecules image tumors using positron-emitters before therapy, while 177Lu provides both therapy and imaging. Current prostate RPT uses standard dosing, but we are developing models to personalize 177Lu-PSMA dosing using imaging data from pre-treatment PET and 177Lu-PSMA SPECT. We are also creating deep learning-assisted models for tailored dosimetry and treatment planning.

Investigators: Moses Wilks, Georges El Fakhri, Thibault Marin