Using ImageJ Software to Estimate Number of

Download PDF
0 downloads 0 Views 122KB Size Report
Comment
New Technique to Count Mosquito Adults: Using ImageJ ..... Major orthogonal dimensions measurement of food grains by machine vision using ImageJ.
New Technique to Count Mosquito Adults: Using ImageJ Software to Estimate Number of Mosquito Adults in a Trap Author(s): Banugopan Kesavaraju and Sammie Dickson Source: Journal of the American Mosquito Control Association, 28(4):330-333. 2012. Published By: The American Mosquito Control Association DOI: http://dx.doi.org/10.2987/12-6254R.1 URL: http://www.bioone.org/doi/full/10.2987/12-6254R.1

BioOne (www.bioone.org) is a nonprofit, online aggregation of core research in the biological, ecological, and environmental sciences. BioOne provides a sustainable online platform for over 170 journals and books published by nonprofit societies, associations, museums, institutions, and presses. Your use of this PDF, the BioOne Web site, and all posted and associated content indicates your acceptance of BioOne’s Terms of Use, available at www.bioone.org/page/ terms_of_use. Usage of BioOne content is strictly limited to personal, educational, and non-commercial use. Commercial inquiries or rights and permissions requests should be directed to the individual publisher as copyright holder.

BioOne sees sustainable scholarly publishing as an inherently collaborative enterprise connecting authors, nonprofit publishers, academic institutions, research libraries, and research funders in the common goal of maximizing access to critical research.

Journal of the American Mosquito Control Association, 28(4):330–333, 2012 Copyright E 2012 by The American Mosquito Control Association, Inc.

OPERATIONAL NOTE NEW TECHNIQUE TO COUNT MOSQUITO ADULTS: USING IMAGEJ SOFTWARE TO ESTIMATE NUMBER OF MOSQUITO ADULTS IN A TRAP BANUGOPAN KESAVARAJU

AND

SAMMIE DICKSON

Salt Lake City Mosquito Abatement District, 2020 N Redwood Road, Salt Lake City, UT 84116 ABSTRACT. A new technique is described here to count mosquitoes using open-source software. We wanted to develop a protocol that would estimate the total number of mosquitoes from a picture using ImageJ. Adult mosquitoes from CO2-baited traps were spread on a tray and photographed. The total number of mosquitoes in a picture was estimated using various calibrations on ImageJ, and results were compared with manual counting to identify the ideal calibration. The average trap count was 1,541, and the average difference between the manual count and the best calibration was 174.11 6 21.59, with 93% correlation. Subsequently, contents of a trap were photographed 5 different times after they were shuffled between each picture to alter the picture pattern of adult mosquitoes. The standard error among variations stayed below 50, indicating limited variation for total count between pictures of the same trap when the pictures were processed through ImageJ. These results indicate the software could be utilized efficiently to estimate total number of mosquitoes from traps. KEY WORDS

ImageJ, mosquito trap, counting mosquitoes

Adult surveillance is an important aspect of mosquito control, and adult mosquito traps play an important role in aiding surveillance. The New Jersey light trap, also referred to as a Model 50 light trap, was developed in 1932 and is still in use in many mosquito abatement programs in the USA (Reisen et al. 1999). Although New Jersey light traps are valuable tools in adult surveillance, apart from mosquitoes they also trap many other insects that are attracted to light, which increases the processing time. Carbon dioxide–baited American Biophysics Corporation (ABC) traps (Clarke Mosquito Control, Roselle, IL) were 1st explored in 1966 and are currently used by many programs (Newhouse et al. 1966, Reisen et al. 1999). The majority of the trap catches in a CO2baited trap are female mosquitoes, and so it reduces sorting time compared to New Jersey light trap catches, especially for arbovirus surveillance. However, the number of mosquitoes trapped in a CO2-baited trap typically is greater than for a New Jersey light trap, which increases the time needed to process the collected mosquitoes. Abundance and species of mosquitoes in a trap are both important for control operations. Although it is difficult to replace human labor for species identifications, attempts have been made in the past to estimate abundances using various techniques. Estimating the number of adults by weighing a trap catch is often suggested as a valuable alternative to human labor, but variations in humidity in the trapping and weighing location could increase the error rate of estimations. Influence of moisture could be reduced by drying the mosquitoes in an oven for a few days, but that will increase the time to estimate the

abundances. Our aim was to develop a system that will efficiently estimate the abundance of mosquito adults in a trap, preferably within a couple of hours of collecting the trap. ImageJ is an open-source image processing software based on Java and is available for download from http://rsbweb.nih.gov/ij/. The software can be installed on any platform (Windows, Mac OS, Mac OS X, and Linux) that has Java installed in the system. ImageJ is a popular software that has users from a wide variety of fields, including medical (Fortin and Battie´ 2012), horticultural (Hill et al. 2005), agricultural (Igathinathane et al. 2009), and entomology (Teale 2009). Sinclair et al (2009) used ImageJ to process X-ray images to understand ice formation during lethal and nonlethal freezing in insects, whereas Mains et al (2008) used it to estimate the number of mosquito eggs on a germination paper. Our goal was to develop a technique to estimate the number of mosquitoes from pictures of mosquito adults collected by a CO2-baited trap using ImageJ software. The ABC traps, baited with CO2, were set out for 24 h within the Salt Lake City area and brought back to the laboratory for analysis between May and October 2011. Location and timing of the trap set up varied, but, in total, 51 traps were used for this experiment. The nets containing trapped specimens were placed in air-tight buckets containing triethylamine to knock down the mosquitoes. The nets were then emptied onto a white photodeveloping tray (20.32 3 25.4 cm; CescoH; B&H Photo Video, New York City, NY). The photodeveloping trays were rubbed with BounceH (Procter & Gamble, Toronto, ON, Canada) fabric

330

DECEMBER 2012

OPERATIONAL NOTE

331

Table 1. Different ImageJ settings used to compare the difference in mosquito counts between manual and ImageJ counts. The ideal setting that produced the least difference between manual and ImageJ count is highlighted in bold. No.

Picture

Threshold

Size

Image J count (average)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8

0,43 0,43 0,43 0,43 0,43 0,43 0,43 0,45 0,50 0,55 0,50 0,43 0,50 0,50 0,50 0,50

43–infinity 45–infinity 50–infinity 40–infinity 30–infinity 20–infinity 55–infinity 43–infinity 43–infinity 43–infinity 100–infinity 100–infinity 45–infinity 50–infinity 53–infinity 60–infinity

1,683.22 1,652.55 1,583.00 1,731.29 1,931.71 2,261.02 1,522.12 1,583.00 1,679.75 1,600.78 1,476.94 1,183.51 1,160.92 1,575.67 1,519.35 1,488.73

softener sheets before the mosquitoes were transferred, which reduced the static electricity and prevented clumping of mosquitoes. Deer flies, which were the only other common insects apart from mosquitoes, were removed. The trays were placed under two table lamps with 100-W fluorescent light bulbs (EcosmartH; Daylight, Home Depot, Atlanta, GA) and photographed with a Canon EOS Rebel T1i digital single-lens reflex camera fitted with an 18–55-mm zoom lens. The camera was placed on a tripod with a horizontal extension, and a remote trigger was used to snap the pictures. The camera was zoomed to show only the developing tray’s base in order to maintain consistency among pictures. Any clumps of adult mosquitoes were separated to create a single layer of adult mosquitoes on the tray, and the average processing time for achieving a single layer was approximately 2 min per trap. After the pictures were snapped, the number of mosquitoes from each trap was manually counted to compare with ImageJ-estimated mosquito counts. ImageJ software allows for varying calibrations in order to adapt to the object of interest in the picture that needs to be estimated. The initial calibration for the ImageJ software, determined according to Mains et al. (2008), was altered multiple times in order to identify the ideal calibration (Table 1). At each calibration, pictures were analyzed through ImageJ to get the estimated number of adult mosquitoes in the picture, and this number was subtracted from the manual count for that picture. In total, 51 pictures were compared for each calibration. The difference between the ImageJ and the manual count for each picture was calculated and averaged by calibration to identify the least average difference (Table 1). After the ideal calibration (#14 in Table 1) was identified,

Difference between manual and ImageJ count 6 SE (average) 235.86 229.35 219.76 253.39 393.52 719.31 220.37 213.88 191.62 229.86 363.92 393.45 182.98 174.11 175.88 193.33

6 6 6 6 6 6 6 6 6 6 6 6 6 ± 6 6

27.87 26.57 27.81 30.82 46.57 72.34 31.01 27.81 27.77 25.48 53.16 77.06 24.01 21.80 22.27 25.34

manual counts were regressed against ImageJ counts for 51 pictures to evaluate correlation and predictability between the manual and Image J count. A two-tailed t-test was also performed to evaluate whether there was a significant difference between the manual count and ImageJ estimation. All the pictures were converted to 8-bit blackand-white mode using the IMAGE: TYPE submenu. The THRESHOLD function, which is under the IMAGE: ADJUST submenu, was used to differentiate between the adult mosquitoes in the picture and the background. The total number of adult mosquitoes in the picture (or the total area where adult mosquitoes were present) was analyzed using ANALYZE: ANALYZE PARTICLES submenu. ANALYZE PARTICLES has several options that can be configured, among which SIZE and CIRCULARITY are most important. SIZE determines the minimum and maximum area (0 to infinity) that should be included in the analysis, and CIRCULARITY (0.00–1.00) determines the shape of the area that should be included. CIRCULARITY was left at the default for all the pictures to include all shapes. EXCLUDE ON EDGES was selected for all the pictures to avoid errors due to perimeters. The accuracy of the ideal calibration to estimate the number of adult mosquitoes in a picture was evaluated by comparing estimations from pictures that were taken after shuffling the tray to rearrange the pattern of distribution of adult mosquitoes on the tray. Contents of 6 CO2baited traps were used for this comparison, and each trap was processed the same way as described before; the only difference is that after the pictures were taken, the tray was shaken to shuffle the distribution of the mosquitoes on the

332

JOURNAL OF THE AMERICAN MOSQUITO CONTROL ASSOCIATION

Fig. 1. (A) Regression between manual count and the ideal ImageJ calibration estimation (#14 in Table 1) for the 51 pictures. (B) Comparison in count between the 1st picture (1) and 5 subsequent pictures (2 through 6) that were redistributed by shuffling the mosquitoes on the tray. Bars in black are the 1st pictures, and bars in gray are the mean 6 SE count of 5 subsequent pictures.

tray, and the tray was rephotographed. The process was repeated 4 times for each trap, making a total of 5 variations for each trap. The initial and 5 variation pictures for each trap were processed through ImageJ for comparison of the accuracy of the calibration to estimate the number of mosquitoes in spite of shuffling to rearrange the pattern of distribution of mosquitoes on the tray. Of 16 different software calibrations tested, the calibration that had the least average difference (174.11 6 21.59) between manual count and ImageJ estimation was with a THRESHOLD 5 0, 50 and SIZE 5 50–infinity (Table 1). The regression analysis based on this calibration indicated that the correlation between the manual count and ImageJ estimation was .93% (Fig. 1A). The t-test indicated that there was no significant difference between the manual count and ImageJ estimation (P 5 0.4703). The initial picture and the average of its 5 variations showed that there were no major differences (SE of variations for each picture 5 50.78, 14.38, 42.85, 15.58, 43.21, 9.38) among them, indicating that

VOL. 28, NO. 4

the ImageJ calibration was able to estimate the number of mosquitoes in a given picture with reasonable accuracy (Fig. 1B). The regression graph shows that traps that had ,2,000 mosquitoes are closer and uniformly distributed in comparison to those that had .2,000 mosquitoes. The adult mosquitoes need to be in a single layer for the ImageJ system to accurately process the pixel area. If the mosquitoes are piled up more than a layer high, then the estimation process could be seriously flawed. We argue that this is not a limitation of the software but rather a limitation of the protocol that we have developed. For traps that have more than 2,000 mosquitoes, a different calibration with a bigger tray size would be ideal. The ImageJ system provides the option of combining the necessary calibrations into a convenient macrofile. So, users could develop calibrations according to the need in their given area and have multiple macrofiles based on the number of mosquitoes in a trap. The calibrations provided in Table 1 could provide a guideline to customize the calibrations according to the requirements of the individual programs. The number of mosquitoes trapped most often is ,250 (e.g., Aedes albopictus (Skuse)) in an urban area but can exceed 15,000 (e.g., Ae. dorsalis (Meigen)) in a rural area, and ImageJ could be configured to work efficiently on either end of the spectrum. Adult surveillance in many abatement programs requires not only the number of mosquitoes but also the species of mosquitoes present in the trap to mount an optimal control operation. Although ImageJ can speed up the process in estimating the total number of mosquitoes in a trap, at the present time, it cannot identify the species of mosquitoes in the trap. A subsampling protocol was developed wherein for each trap, 1 tablespoon full of mosquitoes was sampled and identified. The proportion of species present in a tablespoon was estimated and multiplied by the total number of mosquitoes estimated by ImageJ to get the relative proportion of species in the trap. Further studies are warranted to identify the error rate in estimating the species composition and to develop the optimal subsample relative to the total number of mosquitoes in a trap. Often the predominant mosquitoes in CO2-baited ABC traps within the Salt Lake City area are Culex tarsalis Coq. and Ae. dorsalis. Because of this low diversity, the subsampling (1 tablespoon) was fairly effective, but for areas where the diversity could be high, proper subsampling techniques should be developed for the species ratio estimations to make operational sense. The technique described herein may not work for traps that could potentially trap many groups of insects other than mosquitoes (e.g., New Jersey light traps), as considerable time will be spent in removing those nonmosquito insects prior to

DECEMBER 2012

OPERATIONAL NOTE

taking a picture. In some locations, CO2-baited ABC traps could trap large amounts of other blood-sucking insects, such as eye gnats. ImageJ calibrations could be altered to exclude these small insects from being counted along with mosquitoes, but further research is required to evaluate its efficacy under such a scenario. The ImageJ system can be utilized efficiently to count adult mosquitoes from pictures. The technique was successfully implemented during the 2011 mosquito season in the Salt Lake City area. Each week, .40 traps were set up within the district, and these traps were processed within 4 h of being brought back to the laboratory. The system worked efficiently, and, most often, the time it took to bring the traps back to the laboratory was more than the processing time with ImageJ. We thank Sally Beagley for suggesting the use of fabric softener to reduce static electricity, and Dennis Kiyoguchi, Brad Sorensen, Jessica Hunt, Heidi Fjelstrom, Lon Hokanson, and Brennan Heugly for help with data collection. REFERENCES CITED Fortin M, Battie´ MC. 2012. Quantitative paraspinal muscle measurements: inter-software reliability and agreement using OsiriX and ImageJ. Phys Ther 82:853–864.

333

Hill MG, Mauchline NA, Cate LR, Connolly PG. 2005. A technique for measuring growth rate and survival of armoured scale insects. NZ Plant Protect 58: 288–293. Igathinathane C, Pordesimo LO, Batchelor WD. 2009. Major orthogonal dimensions measurement of food grains by machine vision using ImageJ. Food Res Int 42:76–84. Mains JW, Mercer DR, Dobson SL. 2008. Digital image analysis to estimate numbers of Aedes eggs oviposited in containers. J Am Mosq Control Assoc 24:496–501. Newhouse VF, Chamberlain RW, Johnson JG, Sudia WD. 1966. Use of dry ice to increase mosquito catches of the CDC miniature lighttrap. Mosq News 26:30–35. Reisen WK, Boyce K, Cummings RC, Delgado O, Guttierez A, Meyer RP, Scott TW. 1999. Comparative effectiveness of three adult mosquito sampling methods in habitats representative of four different biomes of California. J Am Mosq Control Assoc 15:24–31. Sinclair BJ, Gibbs AG, Lee W-K, Rajamohan A, Roberts SP. 2009. Synchrotron X-ray visualisation of ice formation in insects during lethal and non-lethal freezing. PLoS ONE 4:e8259. DOI:10.1371/journal. pone.0008259. Teale SA, Letkowski S, Matusick G, Stehman SV, Castello JD. 2009. Quantitative, nondestructive assessment of beech scale (Hemiptera: Cryptococcidae) density using digital image analysis of wax masses. Environ Entomol 38:1235–1240.