Since the scratching assessment is the only way to estimate itching sensation in non-verbal experimental animal, it is utilized in the various research fields. However, current methods depend on human observation, which is laborious, low-throughput, and includes observer-bias. We here aimed to establish an automated scratching detection method of mice using neural network, an artificial intelligence technology which excels in image recognition.
Scratching was elicited by intradermal injection of a pruritogen, lysophosphatidic acid to BALB/c mice and their behavior was recorded with a video camera. Frame images were obtained from video data and classified into two classes: scratching or not. We then trained convolutional recurrent neural network (CRNN) with labeled datasets. Trained CRNN predicted scratching of first-look data with high accuracy (sensitivity: 81.6%, positive predictive rate 87.9%). We confirmed that the number and duration of predicted scratching bouts were comparable to those of human observation. Trained CRNN could also successfully detect scratching evoked by hapten-induced atopic dermatitis (sensitivity: 94.8%, positive predictive rate: 82.1%).
We here established a novel automated scratching detection method using artificial intelligence, which is applicable to the assessment of pathological mouse model.