VCOM Louisiana Research Day Program

Biomedical Research

Naved Salim, BS, OMS-II; Camryn Daidone, BS, OMS-II; Kevin Le, BS, OMS-II; Kristina McLeod van-Amstel, MS, OMS-II; Sarah Voth, PhD; K. Adam Morrow, PhD Edward Via College of Osteopathic Medicine-Louisiana, Monroe, Louisiana 09 ARTIFICIAL INTELLIGENCE IN MICROSCROPIC ANALYSIS

Context: Analysis of microscopic images is a backbone of many areas of research. Digital image processing software such as ImageJ are commonly used to precisely annotate, measure, and analyze images. However, without programming knowledge or Macros, ImageJ is reliant on manual analysis of individual images, a process that is often tedious and leaves data susceptible to human error. Biodock is a new online software that contrasts the traditional manual processing of images as done through ImageJ by allowing users to train Artificial Intelligence (AI) pipelines to identify and quantify objects within an image. Biodock allows users to train pipelines to obtain the desired data and then, once trained, the AI pipelines can be utilized to analyze similar images. Though a potentially less time-intensive and automated approach to image analysis and data collection, the efficacy of Biodock has not been widely tested. Objective: The goal of this study is to determine the efficacy of Biodock AI pipelines in analyzing microscopic images of cells compared to traditional manual analysis by ImageJ. Methods: We tested Biodock’s ability to perform 3 functions compared to ImageJ: measure wound area on scratch wound images,

calculate number of cells on growth curve images, and count number of enclosed loops on network formation images. The sample sizes of images evaluated for scratch wound area, growth curve, and network formation were 126, 45, and 20, respectively. Using ImageJ, wound space area was measured manually using the “freehand selections” function; individual cells and number of enclosed loops within growth curve and network formation images, respectively, were calculated manually using the “multipoint function.” On Biodock these same functions were performed through training three separate AI pipelines, one for each image type, and deploying the trained model to perform these functions on subsequent images. Differences between the means of the data measured from Biodock and ImageJ were calculated using an unpaired two-tailed t-test. We used an alpha value of p = 0.05 to determine statistical significance within a 95% confidence interval. Results: Trained Biodock AI analysis was highly time efficient and resistant to the inherent variability in outcomes that arise via the analyses of an image data set by multiple users. Statistically, there was no significant difference between the Biodock and ImageJ generated image analysis data for scratch wound area,

growth curve, or network formation (p = 0.7024, 0.9517, 0.6800, respectively). Conclusions: These data suggest that Biodock is a promising alternative to ImageJ for quantification of images taken from cell-based functional assays.

21 2023 Via Research Recognition Day