Currently a Professor at the Florida Institute of Technology Debasis Mitra is a Physicsist and a Computer Scientist.

Professional objective of Debasis Mitra is to solve problems in science, engineering, and medicine by applying methematics and computer science. The current theme of his research is to squeeze knowledge nuggets from noisy spatio-temporal data, especially in biomedical images. A long term vision is to understand the molecular biological processes for improving human health. In the past Mitra made contributions in artificial intelligence, studying the qualitative reasoning problems with spatial and temporal constraints.


I learn, along with my students.

For Undergrad advisee's at FIT:
Program plan for CS
Approved Humanities Electives for Spring2015
Approved Science Electives for Spring2015
Restricted Business Electives 2015
(Disclaimer: I received them in 2015. The list is not necessearily accurate/"official"/mandatory, it is rather a suggestive one)

On a: learning contract
Stanford class-central
Stanford coursera
MIT mitx
MIT udacity

Primary question my lab asks is given sequence of images acquired at different points in time and perspective, or spatio-temporal data/image, how much information can we extract from it based on a mopdel? Apart from overall statistical model of noise, we often have extra knowledge about what we are looking for (regularization). Can we embed such knowldge in the processing pipeline to enhance signal to noise ratio? In essence this is combined output of data and qualitative knowledge on the target objective.

Collection of Statistical Validation parameters

Message for prospective students

A talk on Medical imaging

Graduate Students

  • Haoran Chang
  • Valerie Kobzarenko

Alumni (Dissertation/Thesis/Capstone)

  • Gengbo Liu , Ph.D., Genentech Corp., California
  • Haoran Chang, M.S., currently doing Ph.D. with me at FIT
  • Hui Pan, Ph.D., Microsoft Corp.
  • Mahmoud Abdalah, Ph.D., Research Associate, Moffitt Cancer Research Center, U of South Florida, Tampa
  • Stephen Johnson, Ph.D., Dean, East Florida State College, Palm Bay, Florida
  • Chen Shi, M.S., Stanley Black & Decker, Atlanta, GA
  • Bo Li, M.S.
  • Kimberley Day, B.S., at Google Corp.
  • Daniel Eiland, M.S., at Harris Corp.
  • Antall Fernandes, M.S., at Visible Measures Corp, Boston
  • Richard Hoch, M.S., General Dynamics
  • Abhishek, Bannerjee, B.Tech.-IIT, Google, India
  • Florent Launay, M.S., Microsoft Corp.
  • Sung Park, M.S., at Microsoft Corp.
  • Keith Ledig, M.S., at Harris Corp.
  • Gandhali Samanth, M.S., Microsoft Corp.
  • Michael Smith, M.S., (somewhere in New York!)

Alan Bundy's Page on how-to-do research


  • Youngho Seo, Department of Radiology and Biomedical Imaging, UCSF
  • Grant Gullberg & Group, Department of Radiology and Biomedical Imaging, UCSF
    [Previously at: Dept of Biophys. and Integr. Biology, Lawrence Berkeley National Lab, Berkeley, CA, an old Group image
  • Dr. Daniel Rubin, MD., MS., Professor of Biomedical Data Science, Radiology, and Medicine (Biomedical Informatics) and (by courtesy) Ophthalmology, Stanford
  • Marcus Hohlmann, Physics and Space Sc. Department, FIT


Some university rankings

Learning about the laws of nature has always facinated me. So I started my higher education studying physics. I did my first Ph.D. on mathematical physics, working on Lorentz Groups. While pursuing my Ph.D. at IIT I was hired by one of the largest companies, Oil and Natural Gas Commission, in India, my country of origin. I worked as an exploration geophysicist there, on areas like petroleum geology, onshore and offshore field geophysics, and yes, a lot of data processing. I fell in love with the art of programming and became overwhelmed with the idea that the computers could be 'programmed' to think. At this point in my career I decided that I have had enough in oil business and should learn about computers and artificial intelligence more seriously. I joined the University of Louisiana at Lafayette as a Ph.D. student in computer science. 'Time' became once again my area of investigation, as it was during my past graduate study in physics. This time I looked into it from the computational point of view - how to reason with it. After completing my second doctorate I have decided to teach computer science while continuing my journey of delving more and more into 'space' and 'time.'

In the end I became fatigued with working on complete abstractions, and wanted to dip my hands into some real data, as I was doing during my early infatuation with computers. Mathematical or scientific computing started attracting me more. With a sabbatical at Lawrence Berkeley National Lab I was introduced to inverse problems over noisy nuclear imaging data. Medical image processing and management became my staple, but once again, on spatio-temporal data or dynamic images. Thanks to a number of extraordinary collaborators and students, our learning curve in this new area is steady.

Journey continues. As the next stage of the above works, we are now involved with radiomics, asking - how to make prognosis from some 3D med-images of patients? Obviously, this is a machine learning task and our hands are dirty with tools like convolutional neural networks. The field is exploding and our heads whirl with recent new exotic works in the field. However, is there a role for topology in all these? Can studying topological invariances of images enhance learning capbility and efficiency? Are our works in spatio-temporal knowledge representation have some value in this direction?
So, here I am! But, stay tuned for more, in case you are tracking me :)

A disclaimer: I am not the Debasis Mitra, who works on distributed computing, now in Columbia University.



Medical Imaging
Our objective in this project is to transform raw camera data from SPECT (Single Photon Emmission Computer Tomography) and PET (Positron Emmission Tomography) scans into three dimensional or four dimensional views within body. We develop novel reconstruction and optization algorithms for dynamic or time-varying data for better diagnostics, and implement them on high-performance computing systems. We ask if neural network are useful in image reconstruction and if so, are they worth using? We collaborate with imaging scientists, physicists, cardiologists, neurologists, from different labs including those at the University of California San Francisco, the Cardiff University, UK, and the Stanford School of Medicine. [Supported by the National Institute of Health.]

Medical Imaging
We try to provide prognosis of tumor from 3D medical images by using machine learning algorithms. Deep learning is very useful in this context but there are many challenges. Medical data are not as easily available as in the case of 2D images for computer vision research. That is challenge number one! Where to look for the tumor automatically even when we know it is on breast - how does a program know where the breast is in a whole-body image? Semantic segmentation is the challenge number two? We are trying to address these problems with respect to the images' topological invariants. [Supported initially by NIH subcontracts and then, from the Univ. of Calif., San Francisco.]

Muon Tomography
Muons are leptons (weakly interacting elementary particles) that are highly penetrating because of low charge (same as that of electron) and high momentum (about two hundred times heavier than electron). This makes them ideal probe for detecting materials with higher atomic numbers (Z). Steady flux of muons are produced by cosmic ray at upper atmosphere. This type of natural muons can be used as non-invasive probes for charting magma chambers in volcanos, hidden chambers in pyramids, and for detecting Uranium or other high-atmoic number materials in cargo containers for border protection. Our project is related to the last area. Our group collaborates with a High Energy Physics group at FIT. We work on reconstruction algorithms for muon-tomography. [Project was initiated with a support from Dept. of Homeland Security.]


Spatio-temporal Reasoning
A typical Constraint Reasoner checks for satisfiability in an input set of constraints. For a satisfiable set it may also generate a consistent scenario. However, traditionally, nothing is done with an inconsistent set of information. Any user of such a system would often wish that the system were more helpful in debugging an inconsistet situation. We harness this issue in temporal relations where constraints may be qualitative (e.g.,"before or overlaps") or quantitative (e.g., "within 10 to 20 seconds"), and disjunctive in nature.

Another question we have started asking recently (with a colleague) is whether a temporal reasoning approach may be helpful in model-checking of temporal information in designing systems. [Initially supported with a Natl. Sc. Foundation CAREER award.]


Computational Molecular Biology
We try to understand how immune systems evolve after vaccination by applying unsupervised learning techniques.
A protein is a chain of amino acids that forms a complex structure in 3D. Each protein typically has the same structure and its function is often determined by this structure. Thus, proteins are classified according to their strucutures and comparing protein structures is a serious algorithmic business. Phenotypes of organisms, disease functionalities, or drug developments - all need to have understanding of protein structures. We were engaged in studying and developing such structural comparison techniques.


Science or engineering researchers produce copious amounts of data in their experiments. Often management of data becomes unwildy, even when each piece of data sets is not necessarily very large. Typically researchers keep track of data with directory hierarchies and file naming conventions. This becomes extremely cumbersome and expensive in terms of time needed to find out necessary data and may result in repetition of experiments done before. A database for keeping track of metatdata (information on data) is essential. In this project, we studied experiment-related data management. We also addressed the issue of organizing knowledge behind a scientific or engineering experiment and data processing activities. The objective was to help a scientist in answering queries about the experiment or data produced by the experiments. A long term goal of these activities is to understand how creativity can be computationally modeled in scientific works and how such a knowledge management system can debug an experimental procedure. Our application focus was in bio-medical imaging experiments. [This was supported by NIH subcontracts and an NSF grant.]


Potential ideas
Sometimes we explore high-risk questions (in terms of immediate impactful answers) that are very potent in the long run, or at least very interesting! I call them proto-projects. Lack of background and resources to pursue them keep them unfinished, and us a bit frustrated. Here is a collection of some of such recent endeavours.
Temporal reasoning in real life: Can a disjunctive temporal constraint reasoning system help archeology? Or, crime detection?
Hurricane path clustering: Do hurricane paths over the years have some similarity to each other? Do shapes of those paths carry any useful information?
Correlation in Indus Valley script (IVS): How do IVS symbols group together in the available texts? Can we make any semantics out of such groupings? Can we correlate such groupings with those in other similar languages? Are there any spatio-temporal relationships between IVS and other conttemporary scripts from neighboring civilizations, like Sumerian or Linear-B, etc.?
Befor we start such research on a large scale, we need a computer vision system to perform Optical Character Recognition. [A Fulbright-Nehru Award will let us explore this last question.]
Can a generic neural network architecture learn different layers of formal languages in Chomsky hierarchy? What does the NN architecture specializes to after learning each type of language and how does its inner layers of the model look like after training? Are NN good at discovering hidden topologies in a string-space?
In a similar line, can an NN be trained to detect a sequence of numbers, e.g., Fibonacci number? Will it be able to learn a generating function that has pole or singularity? Can it learn a generative mathematical transformation, like inverse FFT?
Neural Network learning over a maze: Path-finding puzzles over mazes may be a great way to explore artificial neural learning. Can a neural network be trained to learn existence of a path in a maze? Or, finding an actual path? How does the learning efficiency relates to the texture or other topological properties of mazes?
For the other past projects please see the my CV


Nuclear Imaging

  • “Reconstruction of 4-D Dynamic SPECT Images From Inconsistent Projections Using a Spline Initialized FADS Algorithm (SIFADS).” Mahmoud Abdalah, Rostyslav Boutchko, Debasis Mitra, Grant T. Gullberg. IEEE Transactions in Medical Imaging, 34(1): 216-228, 2015.
  • “Clustering Initiated Factor Analysis (CIFA) Application for Tissue Classification in Dynamic Brain PET.” Rostyslav Boutchko, Debasis Mitra, Suzane Baker, William Jagust, and Grant T. Gullberg. Journal of Cerebral Blood Flow & Metabolism – Nature, doi:10.1038/jcbfm.2015.69, 2015.
  • “High Performance Fully 3D and 4D Image Reconstruction in SPECT Using a Big Data Analytic Tool Running on a Supercomputer.” Huang S-Y., Lee J.H., Pan H., Boutchko R., Shrestha U., Gullberg G.T., Mitra D., Yao Y., and Seo Y. The 13th International Meeting on Fully Three-Dimensional Image Reconstruction in Radiology and Nuclear Medicine (Fully-3D), Newport, Rhode Island, USA, 2015.
  • “Parallelization of Iterative Reconstruction Algorithms in Multiple Modalities,” D. Mitra, H.Pan, Fares Alhassen, and Youngho Seo. Proc. IEEE Nuclear Science Symposium and Medical Imaging Conference, Seattle, WA, 2014.
  • "SinoCor: Sinogram Level Motion Correction in SPECT", Debasis Mitra, Daniel Eiland, Rostyslav Boutchko, Grant T. Gullberg. Lawrence Berkeley National Laboratory Software Licensee CR-3016, 2013

Muon Tomography

  • "A Volume Clearing Algorithm for Muon Tomography” D. Mitra, K. Day, and M. Hohlmann. Proceedings of IEEE Nuclear Science Symposium and Medical Imaging Conference, Seattle, WA, 2014.
  • "Imaging of high-Z material for nuclear contraband detection with a minimal prototype of a muon tomography station based on GEM detectors." Nuclear Instruments and Methods in Physics Research A, , Kondo Gnanvo, Leonard V. Grosso III, Marcus Hohlman, Judson B. Locke, Amilkar Quintero, and Debasis Mitra. 652 (2011) 16–20
  • “GEANT4 Simulation of a Cosmic Ray Muon Tomography System with Micro-Pattern Gas Detectors for the Detection of High-Z Materials,” "Transactions on Nuclear Science", Vol. 56, No. 3, June 2009, Marcus Hohlmann, Patrick Ford, Kondo Gnanvo, Jennifer Helsby, David Pena, Richard Hoch, and Debasis Mitra. [PDF]
  • “Muon Tomography Algorithms for Nuclear Threat Detection,” "Lecture Notes in Artificial Intelligence" series, Springer Verlag, 2009, Richard Hoch, Debasis Mitra, Marcus Hohlman, and Kondo Gnanvo. [PDF]
  • “Performance Expectations for a Tomography System Using Cosmic Ray Muons and Micro Pattern Gas Detectors for the Detection of Nuclear Contraband,” in Proc. of the IEEE Nucl. Sci. Symp. 2008, Dresden, Germany, pp. 1278-1284, IEEE Cat. CFP08NSS-CDR, ISBN 978-1-4244-2715-4, ISSN 1082-3654, e-Print: arXiv:0812.1007, Kondo Gnanvo, Patrick Ford, Jennifer Helsby, Richard Hoch, Debasis Mitra, and Marcus Hohlman. [PDF]

Spatio-temporal Reasoning

  • “Explanation Generation over Temporal Interval Algebra,” in Qualitative Spatio-Temporal Representation and Reasoning: Trends and Future Directions. Shyamanta M. Hazarika (Editor), Information Science Publishing, ISBN: 1616928689, August 1, 2010.
    Debasis Mitra and Florent Launay. [PDF]
  • “Spatial-reasoning for Agents in Multiple Dimensions,” Journal of Universal Computer Science, vol. 8, no. 8, pp. 774-791, August, 2002
    Debasis Mitra, and Gerard Ligozat. [PDF]
  • “Spatial and Temporal Reasoning: Beyond Allen's Calculus,” Gerard Ligozat, Debasis Mitra and Jean-Francois Condotta, AI Communications (European journal on Artificial Intelligence), Vol. 17, no. 4, pp 223—233, 2004.
  • “Modeling and Reasoning with Star Calculus: An Extended Abstract,” Debasis Mitra, Proc. of Eighth International Symposium on AI and Math, January 2004, Ft. Lauderdale, Florida. http:/
  • “Qualitative Direction Calculi with Arbitrary Granularity,” Jochen Renz and Debasis Mitra, Pacific Rim Conference on AI (PRICAI), 2004.
  • "Characterization of Temporal Sequences in Geophysical Databases," Arie Shoshani, Preston Holland, Janet Jacobsen and Debasis Mitra, Proc. of the Statistical and Scientific Database Management (SSDBM) conference, Sweden, 1996.

Mathematical Physics

  • “The Lorentz group in oscillator realization III - the group SO(3,1),” D. Basu and D. Mitra, Journal of Mathematical Physics, Vol 22, p 946, Am. Inst. Phys., 1981.
  • “The Lorentz group in oscillator realization II - integral transform and matrix elements of SO(2,1),” D. Basu and D. Mitra, Journal of Mathematical Physics, Vol 21, p 636, Am. Inst. Phys., 1980.

Data & Knowledge Management

  • “ReMI: An Object-relational Image Database for Nuclear Medicine Research,” Boutchko, Rostyslav; Fernandes, Antall; Pan, Hui; Abdalah, Mahmoud; Giannakidis, Archontis; Boswell, Martin; Mitra, Debasis; Gullberg, Grant T. Annual Conference of Society of Nuclear Medicine, Vancouver, BC, Canada, 2013.
  • “A Data Management System with Web Interface for Pre-clinical Multi-modality Imaging: ReMI,” Mitra, Debasis ; Pan, Hui; Abdalah, Mahmoud; Boutchko, Rostyslav; Boswell, Martin; and Gullberg, Grant T. The Sixth World Molecular Imaging Congress, Savannah GA., 2013.
  • “Three generations of research in computational creativity and beyond,” in Association for Advancement of Artificial Intelligence Spring Symposium on Creative Intelligent Systems (AAAI Tech Report). (Eds.) Ventura, D., Maher, M. L., and Colton, S.; Stanford, California, March 2008. Debasis Mitra.
  • “Data cleaning and enriched representations for anomaly detection in system calls,” G. Tandon, P. Chan, and D. Mitra. Book Chapter in Machine Learning and Data Mining for Computer Security: Methods and Applications, M. Maloof (editor), Springer, 2005.

Computatinal Molecular Biology

  • “Correlogram-based method for comparing biological sequences,” Debasis Mitra, Gandhali Samant and Kuntal Sengupta. Springer Verlag Lecture Notes on AI, 2006.
For a full list of all publications please see the Google scholar page

Debasis Mitra


Florida Istitute of Technology,
Melbourne, Florida 32901

Office Hours: Monday-Wednesday 2-4pm, Fall 2021

dmitra at dot edu
Phone: 321-674-7737
Fax: 321-674-7046