Project DetailsThis work focused on analyzing and classifying ultrasonic vocalizations (USVs) of mice using deep neural networks from the PANN family, aiming to predict behaviors associated with autism. The primary objective was to evaluate the model's ability to differentiate between mice from the WT (control group) and HT (research group with a genetic mutation linked to autism).
As part of the work, an automated process was developed for USV data analysis, which included:
• Data preparation: Organizing and structuring files, using advanced augmentation tools, and addressing challenges such as significant data imbalance between groups.
• Model adaptation: Modifying PANN family models to meet the specific needs of the project, including architectural adjustments to better align with the acoustic data.
• Results analysis: Deriving insights into model limitations and identifying directions for future improvements.
The results showed that despite the adjustments made, the model struggled to accurately classify between HT and WT groups, particularly due to the significant data imbalance. However, the work provided important insights for future research:
1. The need to improve and balance the dataset.
2. Exploring alternative models or developing custom-designed models.
3. Expanding the analysis and comparing results to previous studies in the field.
Despite the challenges, the project contributed valuable knowledge and tools that can be used in the future to enhance methods in preclinical autism research and ultrasonic vocalization analysis.
