Projects

The recent development of high-temporal resolution functional PET (fPET) introduces an emerging focus on “metabolic connectivity (MC)” , providing a complementary perspective to the hemodynamic-based “functional connectivity (FC)” assessed by fMRI. In this study, we applied a connectivity gradient-based analytical scheme on a resting-state simultaneous fPET-fMRI dataset, aiming to characterize the detailed cortical organization of fPET-derived MC and understand how it differs from the fMRI-derived functional network structures.

In this research project, we employed a novel high-throughput screening system to collect a vast amount of neural activity data (via calcium imaging) from zebrafish exposed to psychotropic drugs. Subsequently, we analyzed the data to explore the influence of psychotropic drugs on the activity across various brain regions, as well as the functional connectivity within the zebrafish’s brain.

In this study, we proposed a multi-regional modular recurrent neural network to simulate the cognitive processes. The model is structured into three different modules: perception, information integration, and decision. Here a transfer learning approach is adopted to investigate generalizability across tasks. After training models on source tasks, we fixed the information integration layers, transferred the models to target tasks, and tested their performance. By comparing the performance of different source-target task pairs, we assessed the similarity between different cognitive tasks.

Nasopharyngeal cancer is a serious tumor in the upper pharynx, requiring careful outlining of key organs for radiotherapy planning. Traditionally, this is done manually on CT images, a slow, labor-intensive process reliant on the physician’s experience. To improve this, we use a pre-trained 3D model and an enhanced TransUNet network to automate segmentation of four key organs with impressive results.