Hong Kong researchers use AI to improve electron microscopy
Scientists at the University of Hong Kong have developed an AI-driven method, EMDiffuse, to improve electron microscopy (EM) imaging. EM has transformed our ability to see intricate cellular details, and volume EM (vEM) extends this capability to three-dimensional nanoscale imaging.
There are currently trade-offs between imaging speed and quality that limit the imaging area and volume in EM. The new method addresses the limitations of speed, quality, and sample size in both traditional EM and vEM. In some tasks, EMDiffuse reduced the data acquisition time by a factor of 36.
EMDiffuse uses diffusion model algorithms to enhance low-quality electron microscopy images, ensuring high-resolution results by using low-quality images as guides. This method prevents blurriness and maintains clarity, which is essential for detailed structural studies.
The method is particularly beneficial for vEM, where current hardware struggles with high-resolution 3D imaging of large samples. EMDiffuse improves vertical image clarity using uniformly detailed training data or self-supervised techniques, allowing for improved depth resolution without specialised data. This flexibility enables researchers to study complex cellular components like mitochondria more effectively.
The research team was led by Dr Haibo Jiang from the Department of Chemistry and Dr Xiaojuan Qi from the Department of Electrical and Electronic Engineering at the university and was published in Nature Communications.
The research marks a step forward in imaging capabilities, potentially transforming investigations into subcellular structures. The team hopes that this technology will unveil new biological mechanisms as it matures, broadening the scope of scientific exploration in biological systems.