Highly accurate and precise determination of mouse mass using computer vision

Abstract

Changes in body mass are key indicators of health in humans and animals and are routinely monitored in animal husbandry and preclinical studies. In rodent studies, the current method of manually weighing the animal on a balance causes at least two issues. First, directly handling the animal induces stress, possibly confounding studies. Second, these data are static, limiting continuous assessment and obscuring rapid changes. A non-invasive, continuous method of monitoring animal mass would have utility in multiple biomedical research areas. We combine computer vision with statistical modeling to demonstrate the feasibility of determining mouse body mass by using video data. Our methods determine mass with a 4.8% error across genetically diverse mouse strains with varied coat colors and masses. This error is low enough to replace manual weighing in most mouse studies. We conclude that visually determining rodent mass enables non-invasive, continuous monitoring, improving preclinical studies and animal welfare.

Keywords: body mass; body weight; computer vision; machine learning; machine vision; mouse diversity; size determination; statistical modeling.

Figure 1

Visual mass determination approach taken to address highly variable segmentation areas observed in our data (A) Flow chart describing the full computational process from open-field video to body mass prediction. (B) Time series of percentage of deviations from the mean segmentation area over 55 min for four individual C57BL/6J, C57BL/6NJ, A/J, and BALB/cJ mice with (mean ± SD) of pixel area reported. Raw frames of the approximate least and greatest segmentation areas are shown to the right. Red bar on the C57BL/6J smallest frame indicates 5 cm.

Figure 2

Adjustment of segmentation area by eccentricity reduces RSD across strains Relative standard deviations (RSDs) of segmentation area $A_{cm}$ (in blue) and the adjusted size $A_e$ (in red) for each of the 62 mouse strains considered in our dataset. Each box shows the interquartile range (IQR) and the black whiskers reach the farthest point within 1.5 IQR of the end of the box.

Figure 3

Model performance (A and B) Comparison of true mass, as measured with a regular scale, and predicted mass, as determined by our process. Data shown are a one-test sample. $R^2$, mean absolute error (MAE), root-mean-square error (RMSE), and mean absolute percentage error (MAPE) are presented. The 45$°$ line indicates a perfect prediction. (A) Prediction quality for the base model (M1) accounting for only segmentation area. (B) Prediction quality for the full model (M6) accounting for true size $A_e$, arena, sex, age, and strain. (C and D) Error and accuracy performance for each model under 50-fold cross-validation. (C) MAE across models. Each box shows the interquartile range (IQR) and the black whiskers reach the farthest point within 1.5 IQR of the end of the box. (D) $R^2$ value across models.

Figure 4

Full model (M6) performance by sex and strain (A) Full model performance faceted by sex with $R^2$, MAE, RMSE, and MAPE values presented for males (in blue) and females (in red). (B) Full model performance faceted by strain (mean ± standard deviation). The four strains from Figure 1B are labeled here, along with the strains with the smallest and the largest mean areas. The test and training data are the same as in Figure 3B for ease of comparison. The full results are presented in Table S1.

Figure 5

Successful longitudinal tracking of individual mouse mass for 22 days in the open field Mouse masses were manually (black) and visually (color) assessed over a 22-day period in the 1-h open field for the same animal. (A) Comparison of performance (mean ± standard error) of previously trained models on prediction of mass on new data from a longitudinal experiment. We found that M6 had the lowest error (MAE, RMSE, and MAPE). (B) A scatterplot comparison of predicted and observed weights of the 16 mice (matching colors in B and C, ${\hat{R}}^2$ = 0.87). Points on the dashed diagonal line indicate perfect predictions. (C) Individual mouse data on each day. Plots of predicted weights (color) to observed weights (black) across each mouse (panels) for each day of the experiment. Graph title panel text color indicates sex of the animal (females are green and males are blue).

Figure 6

Successful visual mouse mass assessment over a multi-day experiment with multiple mice in a home environment (A) Three mice were cohoused with bedding, food, and water for 4 continuous days. We used the TAM to segment each mouse and applied M6 to visually predict mouse mass. (B) We selected 8 video clips and predicted mass visually in every frame using M6 (colored lines). Each mouse’s median predicted mass is shown as a colored horizontal line, and the true masses are shown as two black dashed lines (two mice had equal measured masses). The manual mass was measured at the start of the experiment. (C) MAPE of mass prediction for each of the three mice over the 8 videos. Each box shows the interquartile range (IQR) and the black whiskers reach the farthest point within 1.5 IQR of the end of the box.