Joyita Dutta and Her Team Make Major Impact at SNMMI Conference
Associate Professor Joyita Dutta of the Biomedical Engineering (BME) Department and various members from her Biomedical Imaging and Data Science Lab at UMass Amherst were very conspicuous at the 2023 annual meeting of the Society of Nuclear Medicine and Molecular Imaging (SNMMI) this summer. In addition to the research team’s groundbreaking approach for subtyping Alzheimer’s disease, which Dutta presented at the conference, BME Ph.D. student Vibha Balaji scored a second-place finish in the SNMMI Physics, Instrumentation, and Data Sciences Council Young Investigator Awards competition. BME postdoctoral fellow Tzu-An Song also earned an honorable mention in that same competition.
In addition, SNMMI put out a telling press release about BME postdoc Fan Yang's abstract on the imaging genetics for Alzheimer's subtyping, and Yang also received a 2023 ERF-SNMMI Annual Meeting Travel Award to attend the conference. Beyond all those accomplishments by her trainees, Dutta herself participated in a lively debate in which SNMMI experts exchanged insightful views about the role of artificial intelligence (AI) in nuclear medicine, a debate that triggered a long feature story on the popular website AuntMinnie.com.
The pioneering research presented by Dutta and her students at the SNMMI conference focuses on a cutting-edge approach for subtyping Alzheimer’s disease that promises “broad diagnostic utility” and combines genomic- and tau-PET-imaging data based on a novel clustering framework using “Sparse Canonical Correlation Analysis” (BME’s Joyita Dutta and Colleagues Develop a More Personalized Approach for Subtyping Alzheimer’s Disease). This innovative computational technique for subtyping the genetically complex Alzheimer’s disease is vital for Alzheimer’s-disease patients because different subtypes might also have distinct rates and profiles of cognitive decline.
Meanwhile, Balaji’s second-place entry in the Young Investigators Award competition was titled “Graph-based explanations of tau forecasting for Alzheimer’s disease using graph neural networks.” Her research focuses on pathological “tau tangles,” which serve as key biological markers for the progression of Alzheimer’s disease.
As Balaji’s paper explains, the distinct pattern of pathological tau propagation can be interpreted using graph explainers, which provide an important mapping tool driving the predictability of graph neural networks. As a result of her research, Balaji says that her new method is a promising technique “for shedding light on tau spread and for furthering our understanding of Alzheimer’s disease progression.”
Song’s honorable mention in that same competition was based on his outstanding research abstract titled “Self-supervised PET image denoising using a neighbor-to-neighbor network.”
Even before the annual conference began, the SNMMI distributed a press release about Yang’s auspicious abstract, titled “Genomics- and Image-Guided Subtyping Refines Characterization of Alzheimer’s Disease.” Yang’s abstract dealt with the new computational technique, as described above, which was presented by Dutta at the conference.
As Dutta said about her postdoc’s abstract, “It is a large conference, and they only select a small handful of abstracts for press releases each year. So, it was quite exciting for us that the society chose to underscore the important work we are doing to improve the lives of Alzheimer’s disease patients.”
Finally, Dutta also participated in a well-covered debate about the role of AI in nuclear medicine and its implications for the field, as discussed in detail by experts at the SNMMI annual meeting. Specifically, the experts debated whether or not deep-learning AI algorithms can replace conventional radiomics analysis.
Radiomics, as the AuntMinnie.com article explained, refers to the extraction of “mineable” data from medical images and has traditionally been performed using machine-learning algorithms trained to extract specific imaging features for analysis. The technique has been applied to improve diagnosis, prognostication, and clinical-decision support, with the goal of delivering precision medicine, according to the Radiological Society of North America (RSNA).
Conversely, the RSNA defines deep learning as a class of machine learning that, unlike radiomics, which requires hand-engineered feature extraction from input images, can learn to detect these features automatically.
As Dutta explained in the debate, “There are plenty of examples in the literature that show that, when we use deep-learning-based features as opposed to hand-crafted, traditional, radiomics-based features, we are able to get very high accuracies.” She also said that neural networks have shown the potential to identify the parts of images that are most relevant for the task of most interest.
In other words, according to Dutta, deep-learning approaches have the potential to be less burdensome for physicians. “So, to me, it is a no-brainer,” as Dutta concluded. (October 2023)