TABLE 3.

Complete results summary for the African Goat Improvement Network Image Collection Protocol (AGIN-ICP) developmental stages.

Developmental Stage a , b	Protocol iteration	Test location	Iterative protocol changes	Reasons for changes
1. Developmental	Original	United States	1. Add blue backdrop and stand	1. Blue backdrop and stand to increase color differential of goat to the background
Lead: MJW c , CM a			2. Add 10 foot, or 3-m calibration rope	2. Ten-foot rope ensuring proper camera distance
			3. Timing sequence of demographic, tissue, and image collection	3. Timing to reduce difficulty in sampling, and inconvenience to farmers, animal handlers
2. Field Test (early)	Prototype	Ethiopia, Kenya	1. Add blue drop cloth (ground cloth)	1. Blue drop cloth to increase color differential of goat legs to the background
Lead: MJW, DM b			2. Affix blue backdrop to vehicle, fence, barnetc.	2. Back drop stand is heavy, and inconvenient
			3. Drop small sign on animal’s neck	3. Neck sign only on handler, due to small goats
			4. Add the ‘naked’ or ‘side’ pose	4. To provide an unobscured side view (without the sign) to extract coat color and pattern
			5. Change crayon markers for identifying the pin bones (rear pose) and points of shoulder (front pose) to bright duct tape	5. Bright duct tape to increase visibility of the marks in the images for isolation, and due to melting of crayons in the heat
3. Field Test (late)	Prototype	Uganda, Malawi, Tanzania, Mozambique, Zimbabwe	1. Interactions with multiple field sampling teams showed common questions, confusion, or field issues	1. This stage clarified the need for enhanced protocol documentation, and on accounting for field sampling conditions impacting image quality
Lead: CWM d			2. Iterative image review saw issues not apparent to field sampling teams, i.e., site selection, the need to avoid ‘goat like’ objects (large rocks, other equipment), cleaning the drop cloth to maintain the blue coloretc.	2. Improving field sampling team’s understanding of image processing would improve protocol implementation, leading to the development of the Quick Start Guide showing a high-quality example of each pose - connected to the phenotypic measurement to be extracted from it
			3. AGIN-ICP demo at AGIN II meeting in Uganda (ref AGIN paper)	3. Visualize method and equipment; and a question-and-answer opportunity
4. Field Test (advanced)	Modified	Burundi, Egypt, Mali, Madagascar, Tanzania, Sudan	1. Quick Start Guide produced in English and French	1. Quick Start Guide was designed to accompany the protocol, a one-page (front and back) graphical summary of the full protocol
Lead: PB e			2. AGIN-ICP update and informal training at the AGIN III meeting in Ethiopia (ref AGIN paper)	2, 3. Opportunity for field sampling teams in this stage to ask questions directly, examine sampling kit equipment. This connection to the lead protocol developer provided a personal connection, and a comfort level to contact her for ongoing support
			3. Ongoing support for field sampling teams was provided by email, or phone call as needed
5. Controlled Test	Modified	United States	1. Drop the marking of pin bones (rear pose) and points of shoulder (front pose) with either crayons or tape	1, 2. Image processing confirmed little value from the front pose, pin bones, or point of shoulder
Lead: MJW			2. Drop the front pose	3. Highly controlled collection, with resulting images used to develop the PreciseEdge Image Segmentation algorithm (PE) to extract digital body measurements directly from AGIN-ICP images. This showed AGIN-ICP image measures are highly correlated to real-world animal measurements (Woodward-Greene et al., 2022). The PE algorithm is integrated into user software to return AGIN-ICP digital phenotypic measures in csv, xlsx, and xml, and labeled images for use in machine learning training set data (manuscript in process) for modeling more sophisticated and automated digital phenotype extraction tools
			3. Images collected in this stage were collected in a highly controlled manner, and used to develop and design the image segmentation algorithm and software to accompany the AGIN-ICP for extracting digital phenotypes from the images

^a C. Mukasa led preliminary tests in Uganda, Nigeria.

^b D.M. led a team in Kenya.

^c M.J. Woodward-Greene.

^d C.W. Masiga.

^e P. Boettcher.