To get a better look at brains, Harvard researchers are making microscopes work more like human eyes.<\/p>\n
Until recently, the quest to build high-resolution maps of brains \u2014 otherwise known as \u201cconnectomes\u201d \u2014 was stymied by the slow pace and cost of powerful electron microscopes capable of systematically capturing neuroanatomy down to billionths of a meter.<\/p>\n
But now a team of scientists at Harvard and MIT have found a way to bypass that bottleneck: using machine learning to guide a simpler, less-expensive variety of microscope in real time. The idea is to home in on key details first and minimize time spent on areas of lesser interest \u2014 the same way we might zero in on words on a page instead of margins.<\/p>\n
Researchers say the innovation, known as SmartEM, will speed scanning sevenfold and open the field of connectomics to a broader research community, boosting our understanding of brain function and behavior.\u00a0<\/p>\n
\u201cSmartEM has the potential to turn connectomics into a benchtop tool.\u201d<\/p>\n
Aravinthan Samuel<\/p>\n
\u201cSmartEM has the potential to turn connectomics into a benchtop tool,\u201d said\u00a0Professor Aravinthan Samuel<\/a>, a researcher in the Department of Physics and Center for Brain Science and one of the senior authors of a\u00a0new paper<\/a> published in Nature Methods. \u201cOur goal is to democratize connectomics. If you can make the relatively common single-beam scanning electron microscope more intelligent, it can run an order of magnitude faster. With foreseeable improvements, a single-beam microscope with SmartEM capability can reach the performance of a very expensive and rare machine.\u201d<\/p>\n The method is the product of a five-year collaboration between researchers at Harvard, MIT, Johns Hopkins Applied Physics Laboratory, and microscope manufacturer Thermo Fisher Scientific.<\/p>\n In December, the same journal proclaimed electron microscopy-based connectomics its \u201cMethod of the Year<\/a>\u201d for 2025 and cited SmartEM an example of cutting-edge innovation.\u00a0<\/p>\n SmartEM marks a new advance in the decades-long quest to create \u201cwiring diagrams\u201d of brains from across the animal kingdom, from worms to fruit flies to humans.\u00a0<\/p>\n For example, two years ago Harvard researchers\u00a0published<\/a>\u00a0the\u00a0first nanoscale map of one cubic millimeter of human brain<\/a>. Packed into that poppy-seed-sized sample were 150 million synapses, 57,000 cells, 230 millimeters of blood vessels, and a wondrous diversity of structures never seen before.\u00a0<\/p>\n Researchers elsewhere have completed connectomes for the fruit fly and zebrafish. The next grand challenge is one for the mouse.<\/p>\n To build these maps, scientists have relied on a technique known as serial-section electron microscopy. It entails shaving samples of brain tissue into thousands of ultra-thin sections, which are then scanned and imaged by powerful electron microscopes.<\/p>\n Next the images are stacked on top of each other to create 3D digital replicas. For example, that one cubic millimeter of human brain tissue published in 2023 was sliced into more than 5,000 sections, each thinner than one-thousandth of a human hair.\u00a0<\/p>\n These endeavors pose monumental technical hurdles for both capturing the images and processing the data.<\/p>\n Until recently, connectomics had been the exclusive purview of a small number of researchers and institutions that can afford multimillion-dollar hardware such as high-throughput electron microscopes with up to 91 beams.\u00a0<\/p>\n With growing demand to generate brain maps of many species, one obvious way to push forward connectomics is to recruit more microscopes \u2014 particularly single-beam electron microscopes, which are widely available at research institutions around the world.<\/p>\n Their speed largely is a function of the \u201cdwell time\u201d that the beam devotes to each pixel. In the standard approach, specimens are scanned with the same high resolution for all pixels.<\/p>\n \u201cWe usually shoot the picture first and then we aim.\u201d<\/p>\n Jeremy R. Knowles<\/p>\n \u201cWe usually shoot the picture first and then we aim,\u201d said Science Dean\u00a0Jeff Lichtman<\/a>, Jeremy R. Knowles Professor of Molecular and Cellular Biology and Santiago Ram\u00f3n y Cajal Professor of Arts and Sciences, a co-author of the new paper. \u201cFrom the data that you\u2019ve shot, you then look at the particular things that you find interesting. And that\u2019s not the way our eyes work.\u201d<\/p>\n Instead, people focus their attention on key details such as the eyes and mouths of other faces or a fly on the wall instead of the larger white background.<\/p>\n