Professor Patrick Brennan (the University of Sydney)
Title: Test set infrastructures: improving cancer detection
Abstract: Early diagnosis of breast cancer results in a 97% survival rate. However, to achieve this survival rate and even more importantly to achieve zero deaths from breast cancer in the near future, we must significantly reduce the 20-30% of breast cancers that fail to be diagnosed. In 2011-2012 the National Breast Cancer Foundation, the Royal Australian and New Zealand College of Radiologists and the Australian Government provided funding to allow the University of Sydney to implement and establish BREAST (BreastScreen Reader Assessment Strategy), a mammographic test set-based research infrastructure focused towards transforming breast cancer diagnosis. To date, the BREAST platform and related innovations have helped to better our understanding of:
- types of missed cancers;
- impact of novel imaging technologies; and
- radiologist characteristics and practices that promote accurate diagnoses.
With 800,000 mammography studies performed in Australia annually and 1 million new breast cancer cases being reported each year globally, the impact of radiologic misdiagnoses on public health is a hugely important issue. In 2011, a study showed that 44% of lesions in test cases were missed by 116 Australian and New Zealand breast imaging readers. In addition, the median level of sensitivity was below 70%. It were data such as these that led to the development of BREAST. It is not that Australian readers perform better or worse than anywhere else, although recent evidence shows the former, we now had established a system of research that highlighted varying levels of performance between individuals and regions.
Since its inception, up to 80% of Australia’s breast imaging radiologists and trainees have engaged with test set approaches resulting in a million data entries. Through BREAST workshops and the online platform, up to 720 individuals world-wide have engaged in experiments – much more than the typical sample size in other studies anywhere in the world. Test sets represent an effective method of engaging enough radiologists to help identify causal agents for human error, and a platform for testing potential solutions. At the time of writing, our approach has been adopted in Australia, New Zealand, Singapore, China, Vietnam, and Mongolia. Data provided to date demonstrated that involvement in our approach improves radiologic performance by over 30%.
Dr Hamish Brown (the University of Melbourne)
Title: From "blobology" to Angstrom resolution molecular structures, progress in cryogenic electron microscopy in the past few decades
Abstract: The transmission electron microscope (TEM), invented by Ernst Ruska in the 1930s, showed early promise for biological applications by allowing viruses, smaller than the resolution limit of that era's light microscopes, to be imaged for the first time. In the ensuing decades the resolution of TEMs approached a few Angstroms, at least for material's science samples, allowing crystal lattices to be routinely imaged. For a time, reaching similar resolutions in biology proved much more difficult due to sample related constraints. A high-energy electron beam requires that a TEM column be kept at ultra-high vacuum, far from the aqueous environment biological molecules prefer, and biological molecules deteriorate rapidly under the electron beam. Cryogenic electron microscopy (cryo-EM) has proved to be an effective work around for both of these problems. By preparing a thin miniscus of solution on an electron microscopy grid and freezing the grid to < -160°C in a few nanoseconds via the method of plunge-freezing, biological samples can be preserved within vitreous ice in a near-native state. Vitreous ice is stable under ultra-high vacuum and slows electron beam damage of biological molecules by preventing the diffusion of free radicals resulting from electron beam radiolysis. A vitrified protein will still only tolerate a few 10s of electrons/Å2, leading to raw images with very low signal to noise, but by imaging many thousands of copies of the same protein in different orientations modern Bayesian reconstruction algorithms can reconstruct three-dimensional models of these particles at a few Angstrom resolution or better. This talk will review the developments that make this incredible level of resolution possible from the development of plunge freezing, to the advent of "direct" electron counting detectors and Bayesian reconstruction techniques.