HKU Uses AI to Overcome Challenges in Electron Microscopy

Inspired by recent advancements in AI-powered image generation models, particularly the development of advanced diffusion models, a research team led by two professors from The University of Hong Kong (HKU) has introduced a suite of diffusion model-based algorithms called EMDiffuse. This innovative solution aims to enhance imaging capabilities and address the trade-offs faced by both EM and vEM. Their findings have been published in Nature Communications.

Electron microscopy (EM) has significantly advanced the ability to visualise intricate cellular details. The progression to three-dimensional electron microscopy (vEM) has further expanded the capability to image at nanoscale resolution. However, challenges related to imaging speed, quality, and sample size continue to limit the achievable imaging area and volume. Concurrently, artificial intelligence is playing a pivotal role in scientific progress, driving breakthroughs and serving as an essential tool in various scientific domains.

For conventional two-dimensional EM, EMDiffuse excels at restoring realistic, high-quality visuals with high-resolution ultrastructural details, even from noisy or low-resolution inputs. Unlike other deep learning-based denoising or super-resolution methods, EMDiffuse employs a unique approach by sampling the solution from the target distribution. It integrates low-quality images as a condition or constraint at each step of its diffusion-based process to ensure the accuracy of the generated structure. This means the low-quality input actively guides and shapes the restoration rather than merely serving as a starting point. The diffusion model effectively prevents blurriness, maintaining a resolution comparable to ground truth, which is crucial for detailed ultrastructural studies. Furthermore, EMDiffuse’s generalisability and transferability allow its application to various datasets directly or with minimal fine-tuning using just one pair of training images.

In the realm of vEM, existing hardware often struggles to capture high-resolution three-dimensional images of large samples, particularly in the depth (or ‘z-direction’), making it challenging to fully study the 3D structure of critical cell components like mitochondria and the endoplasmic reticulum (ER).

EMDiffuse addresses this limitation with two flexible approaches. It can utilise ‘isotropic’ training data – 3D image datasets with uniform, high resolution in all dimensions – to enhance the axial resolution of other 3D data. Alternatively, EMDiffuse can analyse existing 3D images and improve their depth resolution through self-supervised techniques without requiring specialised training data. This versatility allows EMDiffuse to enhance the quality and utility of 3D electron microscopy data across different research applications.

The restored volumes demonstrate exceptional accuracy in studying ultrastructural details, such as mitochondrial cristae and interactions between mitochondria and the ER, which are challenging to observe in original anisotropic volumes. Since EMDiffuse does not require isotropic training data, it can be directly applied to any existing anisotropic volume to improve its axial resolution.

EMDiffuse represents a significant advancement in the imaging capabilities of both EM and vEM, enhancing image quality and axial resolution of the data produced. This foundation sets the stage for further development and acceleration of the EMDiffuse algorithm, paving the way for in-depth investigations into the intricate subcellular nanoscale ultrastructure within large biological systems. As this AI-powered imaging technology matures, it is anticipated to enable researchers to uncover previously undiscovered operational mechanisms within biological systems.

By leveraging advanced diffusion models, the team from HKU has provided a powerful tool that addresses key challenges in electron microscopy. EMDiffuse not only improves the visual quality and resolution of EM images but also enhances the depth resolution of 3D electron microscopy data. This innovation underscores the potential of AI to drive significant progress in scientific imaging, offering new opportunities for detailed exploration of cellular structures and their functions. The work of this team illustrates the transformative impact of integrating AI with traditional imaging technologies, setting a new standard for future research in the field.