Fig. 3: Training and inference procedure of the swallow detection method. | Communications Medicine

Fig. 3: Training and inference procedure of the swallow detection method.

From: A deep learning-based approach to enhance accuracy and feasibility of long-term high-resolution manometry examinations

Fig. 3

During the training stage, the classifier is trained to classify preprocessed manometry sequences into two categories (swallows and non-swallows). During inference, a rolling window is moved across the full LTHRM sequence. Each window is preprocessed similarly to the training sequences and then passed to the classifier for inference. The resulting list of output classes and confidences are then processed to identify all groups of swallows. For each swallow event, the predicted swallow start is returned.

Back to article page