Project DetailsSpinal and back surgeries carry the risk of causing damage to the patient's neural function. To mitigate this risk, SSEP signals are measured in real time during surgery to detect potential neural impairments as early as possible. The goal of this project was to develop a system based on deep learning and signal processing that classifies SSEP signals into four categories, achieving a Recall of 80% for the category of significant drop in neural function. The signals were fed into a deep neural network in a time-sampled format, and the model's architecture was based on convolutional layers to exploit the temporal correlation of the signals. Additionally, we developed a preprocessing unit for the signals, incorporating tools from the field of signal processing, such as filtering, decimation, and STFT, to enhance the quality of the input data and provide the network with additional signal features.
Throughout the project, we addressed significant challenges, including inconsistent labeling of samples and an imbalanced class distribution. Using data augmentations and advanced regularization techniques from the field of deep learning, we successfully developed a high-performance model that meets the success criteria defined by the company.
