Teaching models to cope with messy medical data
Peer-Reviewed Publication
Updates every hour. Last Updated: 6-Apr-2026 11:16 ET (6-Apr-2026 15:16 GMT/UTC)
When labelled scans are scarce and hospitals collect images in different ways, a new training recipe developed by SUTD researchers helps segmentation AI keep its bearings across domains without needing more annotations.
Professor Zaifa Shi's team at Xiamen University developed an ultra-high temperature flash vacuum pyrolysis (UT-FVP) device to form giant fullerenes from single-carbon molecules within a short time (15 s) at extremely high temperatures (∽3000 ℃). Due to the strong intermolecular forces between giant fullerene molecules and soot, traditional ultrasonic or Soxhlet extraction methods cannot separate most giant fullerenes from soot in toluene. To overcome these strong intermolecular forces, two separation techniques—mechanical grinding and sublimation—were optimized to separate the giant fullerenes from the pyrolysis products, and laser desorption/ionization time-of-flight mass spectrometry (LDI-TOFMS) was used for comprehensive and thorough detection. These methods extended the mass distribution of synthesized giant fullerenes to 2760 Da (greater than C230). Notably, the separation technology can also recover giant fullerenes that have long been neglected due to incomplete separation in flame and arc discharge methods. This separation strategy has broad applicability in the synthesis of giant fullerenes, providing a new perspective for the synthesis and utilization of these carbon materials. The article was published as an open access research article in CCS Chemistry, the flagship journal of the Chinese Chemical Society.
An array of seismic sensors deployed to capture aftershocks from the 2018 magnitude 7.1 Anchorage earthquake also collected distinctive signals from hundreds of flights crossing over Alaska.