- Andres Rios - dar2196
- Carlo Provinciali - cp2984
- Paridhi Singh - ps3060
- Jane Dong - zd2221
- Xinyuan Cao - xc2461
- Akhil Punia - ap3774
Tal Danino, Assistant Professor, Biomedical Engineering, Fu Foundation School of Engineering and Applied Science
Email:td2506@columbia.edu
Lab: http://daninolab.nyc/
Dhruba Deb, Ph.D. Postdoctoral Research Scientist
Email: dhrubadeb@gmail.com
https://dhrudeb.com/
Svebor Karaman, Postdoc, Electrical Engineering
Email: sk4089@columbia.edu
Sining Chen
Email: sc4549@columbia.edu
Bacteria-related illnesses – caused by strains of Escherichia coli, Salmonella enterica, Listeria monocytogenes, Vibrio cholerae, Corynebacterium diptheriae, methicillin-resistant Staphylococcus aureus (MRSA), and Pseudomonas aeruginosa—are responsible for approximately 5 million deaths per year worldwide. Many of these infectious diseases are not only prevalent in the developing world, but are becoming more common in the developed world with increasing instances of food contamination, hospital-acquired infections, and bioterrorism.
Identification and monitoring of bacteria outbreaks and antimicrobial resistance is critical for tracking health developments and is recommended by the WHO. Current microbial identification approaches can be extraordinarily time consuming, laborious, and expensive. Specifically, in the case of gold-standard clinical microbiology assays, such as staining and microscopy following culture on solid media in Petri dishes, the assistance of experts for morphological identification is required. Furthermore, these identification methods are often sample destructive. Thus, the development of a rapid, automated process to identify and characterize bacterial species from environmental and clinical samples would be a major health innovation.
The aim of this proposal is to develop a machine-learning based approach for the identification and characterization of bacteria via analysis of photographs of bacteria colonies from environmental samples grown on solid media in Petri dishes. The aim would be to build towards machine learning algorithm which could classify colonies on a plate into species, based on aspects such as colony morphology, color, and other features.
A chapter written by faculty sponsors with useful concepts and bio terminologies can be found here
Images of bacterial colonies from soil samples. All in jpeg format: Scanned images of Petri dishes with well-separated colonies from a single species per plate (for training) and scanned images of Petri dishes, captured by Epson scanner at resolution ranging from 400 to 3200 dpi.
Additional Petri dishes from different species on the same plate (goal is to classify these).
Images can be downloaded from this link (200 images, about 100MB size).
The goal of this project is to build an algorithm that analyzes an image of a Petri dish with colonies from different bacterial species and classifies them by species.
One such deep learning classifier has already been developed to classify colonies from two species, K. pneumoniae and E. coli, which could be extended for the more varied dataset. However, there are a variety of approaches and the research goal of this project is to determine which types of algorithms for this task can perform best.
Project Plan can be found and updated here
Meeting minutes can be found and uploaded here
Q&A can be found and posted here