Johns Hopkins Biomedical Engineering primary faculty
Michael I. Miller, PhD
Herschel and Ruth Seder Professor and
University Gilman Scholar
Director, Center for Imaging Science
Co-Director, Kavli Neuroscience Discovery Institute
Office: Clark Hall 301
Lab: Center for Imaging Science
B.S.E.E. State University of NY at Stony Brook, 1976
M.S.E.E., The Johns Hopkins University, 1978
Ph.D., B.M.E., The Johns Hopkins University, 1983
Miller did his earliest doctoral work in Neuroscience on neural codes in the Auditory system in the Neural Encoding Laboratory at Johns Hopkins University during the Johnson, Mountcastle, Sachs and Young era. Miller focussed on rate-timing population codes of complex, speech features including voice-pitch and consonant-vowel syllables  encoded in the discharge patterns across the primary auditory-nerve. These neural codes were part of the basis for the discussions at the 1982 New York Academy of Science meeting on efficacy and timeliness of Cochlear implants.
Miller's impact in the field of statistical iterative image reconstruction for Medical imaging and brain mapping was at Washington University working with Donald L. Snyder on time-of-flight positron emission tomography (PET) systems being instrumented by Michel Ter-Pogossian. Miller's contribution was to stabilize likelihood-estimators of radioactive tracer intensities via the method-of-sieves . This became one of the main approaches for controlling noise artifacts in the Shepp-Vardi algorithm in the context of low count, time-of-flight emission tomography.
During the 90's, Miller joined the Pattern Theory group at Brown University to work with Ulf Grenander on problems in image analysis with Markov random field models. Grenander and Miller collaborated for two decades working on human shape and form during which time Miller remained a visiting Professor within the Pattern Theory group of the Division of Applied Mathematics at Brown University. Grenander and Miller published influential papers on Computational anatomy as a formal theory of human shape and form.  By 2005, the Computational anatomy framework establishing high-dimensional brain mapping via diffeomorphisms at the morphological scale of MRI had become the de facto standard for cross-section analyses of populations studied via 1mm MRI. Codes now exist for diffeomorphic template or atlas mapping, including ANTS, DARTEL, DEMONS,, LDDMM, and StationaryLDDMM, all actively used codes for constructing correspondences between coordinate systems based on sparse features and dense images.
In 1998, Mumford while in Paris encouraged the collaboration on Computational Anatomy and shape between Miller and the Ecole normale superieure de Cachan group of Trouveand Younes which continues to date. They have published continually, supporting numerous exchanges between the CMLA Center for Mathematical studies, and The Johns Hopkins University Center for Imaging Science. Noteworthy was publication of the geodesic equations generalizing the Euler equation for hydrodynamics supporting localized scale-compressibility, the conservation of momentum law for shape momentum, and the Hamiltonian formalism.
During these years, Miller and Csernansky  had developed a long-term research effort on neuroanatomical phenotyping of Alzheimer's disease, Schizophrenia and mood disorder. In 2005, they published with John Morris an early work on predicting conversion to Alzheimer's disease based on clinically available MRI measurements using the diffeomorphometry technologies. In 2009, the Johns Hopkins University BIOCARD project was initiated, led by Marilyn Albert, to study preclinical Alzheimer's disease. In 2014, the BIOCARD team with Younes demonstrated that the original Braak staging of earliest change associated to the entorhinal cortex in the medial temporal lobe could be demonstrated via diffeomorphometry methods in the population of clinical MRI's, and subsequently that this could be measured via MRI in clinical populations upwards of 10 years before clinical symptom. This has the potential to impact clinical treatment of the disease.
Medical image understanding is where speech and language modeling was in 1980. Back then there were few speech, language and text understanding systems. Most of the greatest work had been focusing on the digital and acoustic stream; little progress had been made on the linguistics and structural representation of the language itself. This has all changed, with the entire focus moving away from the acoustic models to the higher level representations of information. Analogously, today in imaging there is a plethora of magnificent imaging devices of all kinds, at all scales, at all prices. However, there is little in the way of image understanding systems, systems which bring value not from the measured picture but from knowledge bases in the world. Think about it, in the context of Medical image reconstruction, could it really be that there is more information from a single MRI scan than in all of the books that Neuroscientists and Radiologists have catalogued since the Renaissance era of da Vinci and Michelangelo? Why then are there no examples of medical imaging devices which inject such information into their modality.
The difficulty has been that biological shape is exquisitely variable, with few methods for computing or measuring and computing the closeness of the geometric representation of normal and abnormal anatomical and biological structures. Without such “parsing algorithms” there are no structured ways to build models of real-world knowledge, or for inserting book knowledge into the medical image understanding algorithms. This is all changing now with the emergence of Computational Anatomy in the field of Medical Imaging. Computational Anatomy, like Computational Linguistics, allows for the semantic representation of brains into their parts and connectivities.
Biomedical engineering students Manisha Aggarwal, Kwame Kutten, Daniel Tward, Yajing Zhang, are currently engaged in the characterization of the human populations, studying the functional and structural characteristics of the human brain. Disease populations are being studied associated to Huntington’s disease, dementia, bipolar disorder, schizophrenia and epilepsy. We are currently hard at work building brain clouds for delivering image understanding algorithms associated to neuropsychiatric illness.
Michael I. Miller, Andreia V. Faria, Kenichi Oishi, and Susumu Mori, "High-throughput neuro-imaging informatics," Front Neuroinform. 2013; 7: 31.
Susumu Mori, Kenichi Oishi, Andreia V. Faria, Michael I. Miller, "Atlas-Based Neuroinformatics via MRI: Harnessing Information from Past Clinical Cases and Quantitative Image Analysis for Patient Care ," Annu Rev Biomed Eng. Author manuscript; available in PMC 2013 July 23.
D. J. Tward, J. Ma, M. I. Miller, L. Younes, "Robust Diffeomorphic mapping via Geodesically Controlled Active Shapes," International Journal of Biomedical Imaging, N pages (2013).
Y. Zhang J. Zhang, et al., Atlas Guided Tract Reconstruction for Automated Examination of the White Matter Anatomy, Neuroimage, 2010, vol. 52, no. 4, 1289-1301.M.I. Miller, Computational Anatomy: shape, growth and atophy comparison via diffeomorphisms, Neuroimage, 2004, vol. 23, Suppl 1: S19-33.
U. Grenander and M. I. Miller, "Computational Anatomy: An Emerging Discipline," Quart. App. Math., 1998, vol. 56, pp. 617-694.