Figure 6. Application of the Selfee (Self-supervised Features Extraction) pipeline to mice open-field test (OFT) videos.

(A) Image preprocessing for Selfee. The area of the behavior chamber was cropped, and the background was extracted. Illumination normalization was performed after background subtraction. This preprocessing could be skipped if the background was consistent in each video, as our pipeline for fly videos (dashed lines). (B) Anomaly detection of mice OFT videos after chronic immobilization stress (CIS) experiences. Only 12 frames (red points, indicated by arrows) were detected based on a threshold constructed with control mice (the blue region), and anomaly scores were slightly higher than the threshold. (C) Dynamic time warping (DTW) analysis of mice OFT videos after CIS experiences. The dynamic difference between control groups and CIS groups was visualized, and positive values indicated a delay of the reference (control groups). Results from Selfee features and animal positions were similar (red and blue lines, respectively). (D) Autoregressive hidden Markov model (AR-HMM) analysis of mice OFT videos after CIS experiences. Principal component analysis (PCA) visualization of state usages of mice in control groups (n=17, blue points) and CIS groups (n=17, red points). Same as Figure 5B.

Figure 6—figure supplement 1. Anomaly detection of chronic immobilization stress (CIS) mice.

(A) Anomaly scores of CIS mice with half of mice in the control groups as references and another half as negative controls. The blue region indicates the max anomaly score of negative controls; blue dots indicate anomaly scores fall into the blue region; red dots indicate detected anomalies (12 points in total). Detected anomalies were only of scores slightly higher than the threshold. (B) Twelve anomalous frames. These frames contained mice near walls with their head facing the center of the arena. All frames seemed normal.

Figure 6—figure supplement 2. Dynamic time warping (DTW) analysis of chronic immobilization stress (CIS) and control mice.

(A) DTW analysis using Selfee (Self-supervised Features Extraction) features of mice in CIS groups with control mice as reference. The red line represents CIS groups, and the blue line represents control groups. The red line is slightly above the blue line. (B) DTW analysis using mice distances to two nearest walls of mice in CIS groups with control mice as reference. The red line represents CIS groups, and the blue line represents control groups. The red line is slightly above the blue line.