Detection of Seizures and Their Similarity in Dogs Using Deep Learning Technique

Seizures are a common neurological disorder in dogs, potentially leading to life-threatening conditions. Early detection is crucial for timely intervention. This research presents a Bidirectional LSTM deep learning method for the detection of seizures in dogs. Our methodology utilizes DeepLabCut with ResNet50 for pose estimation, identifying key anatomical points on dogs through supervised learning. These keypoint coordinates (x, y) are then fed into a Bidirectional LSTM network to classify dog behaviors as seizures or non-seizure activities (including sitting, lying down, walking, running, and jumping). Experimental results demonstrate that the Bidirectional LSTM model achieves a significantly higher accuracy of 90% while maintaining a comparable recall of 85% in seizure detection, outperforming traditional methods like SVM, which yielded approximately 82.5% accuracy and the same 85% recall. This technology has diverse applications, including home-based seizure monitoring systems, alleviating owner anxiety, enabling early seizure detection for timely treatment, and enhancing overall dog care and management. For veterinarians, this system enhances diagnostic capabilities, improves workflow efficiency, and facilitates more accurate treatment monitoring, leading to better patient outcomes and improved quality of life for dogs with seizures.