172858 A pattern recognition computer program to identify Mycobacterium tuberculosis in a microscopic-observation drug-susceptibility (MODS) culture

Tuesday, October 28, 2008

Alicia Katherine Alva Mantari , Universidad Peruana Cayetano Heredia, Lima, Peru
Mirko Zimic, PhD , Universidad Peruana Cayetano Heredia, Lima, Peru
David A. J. Moore, MD, MSc , Wellcome Centre for Clinical Tropical Medicine, Imperial College London, London, United Kingdom
Robert H. Gilman , Department of International Health, Bloomberg School of Public Health, Johns Hopkins University, Baltimore, MD
Mark F. Brady, MMS , Warren Alpert School of Medicine, Brown University, Providence, RI
BACKGROUND: An international resurgence of tuberculosis (TB) is being fueled by the neglect of public health systems, increasing AIDS rates, and the emergence of multi-drug resistant TB (MDRTB). Diagnosis of TB infection and determination of drug susceptibility are important because appropriate drug therapy cures the individual and prevents infection of others; however, diagnosing TB accurately and quickly has been a major challenge. The microscopic-observation drug-susceptibility (MODS) test is a direct observation culture method that simultaneously yields drug susceptibility and is an improvement on current diagnostics in terms of accuracy, speed, and cost. Scaling up this test internationally is difficult because well-trained laboratory technicians are necessary to read the test results.

PURPOSE: To create an image pattern recognition computer program that can correctly identify Mycobacterium tuberculosis (MTB) growth in MODS cultures using only free computer software programs.

METHODS: MODS MTB cultures and non-MTB cultures were digitally photographed and put through a series of image processing procedures. Logistic regression was used to make an algorithm that identifies specific morphological characteristics and geometrical relationships typical to MTB. This algorithm was then challenged to identify a series of MTB and non-MTB digital micrographs.

RESULTS: The pattern recognition algorithm demonstrated a high sensitivity and specificity for identifying MTB growth in MODS compared to atypical mycobacteria and other bacterial growth.

DISCUSSION: Pattern recognition computer programs show promise for helping to scale-up worldwide access to MODS testing for TB and MDRTB. This pattern recognition algorithm may be useful for removing human error, quality control, a teaching tool, creating a high-throughput system, or a way for remote laboratories without experienced laboratory technicians to accurately diagnose tuberculosis. It may be possible to increase access to other microscope diagnostics by following a similar approach, helping to quickly spread health technologies across the borders that infectious diseases do not acknowledge.

Learning Objectives:
Participants will learn how pattern recognition computer programs are constructed and how they can be applied to microscope diagnostics for the purpose of quality control, training laboratory technicians, and creating high-throughput diagnostic systems. This information will be presented from the perspective of international scale-up of the MODS (microscopic-observation drug-susceptibility) diagnostic for tuberculosis. By the end of the presentation, participants will be able to define “pattern recognition,” describe the advantages and disadvantages of pattern recognition versus human laboratory technicians, recognize applications for pattern recognition algorithms, and describe the three steps to creating such an algorithm. Specifically, participants will be able to: 1. List 3 advantages and 3 disadvantages of pattern recognition computer algorithms for laboratory diagnostics. 2. Apply the information learned to identify an application of pattern recognition algorithms in their own work or research. 3. Articulate how they would go about constructing a pattern recognition algorithm for the application they identified.

Keywords: Information Technology, Infectious Diseases

Presenting author's disclosure statement:

Qualified on the content I am responsible for because: Assisted with data analysis and interpretation.
Any relevant financial relationships? No

I agree to comply with the American Public Health Association Conflict of Interest and Commercial Support Guidelines, and to disclose to the participants any off-label or experimental uses of a commercial product or service discussed in my presentation.