Fig 1 - uploaded by Alex Deakyne
Content may be subject to copyright.
Source publication
Recent advancements in deep learning have led to the possibility of increased performance in computer vision tools. A major development has been the usage of Convolutional Neural Networks (CNN) for automatically detecting features within a given image. Architectures such as YOLO1 have obtained incredibly high performances for the real-time detectio...
Contexts in source publication
Context 1
... dataset was split into training and validation sets and all slices were obtained from the transverse plane and resized to 128x128 pixels. A sample slice from one such dataset is depicted in Figure 1a. The 'ground truth bounding boxes' for each slice were created by calculating the relative coordinates of the smallest box that contains the entirety of the left atrium. ...
Context 2
... 'ground truth bounding boxes' for each slice were created by calculating the relative coordinates of the smallest box that contains the entirety of the left atrium. This was done by utilizing the corresponding mask of the left atria for every slice in the dataset ( Figure 1b). is used as the input to the network. ...
Context 3
... dataset was split into training and validation sets and all slices were obtained from the transverse plane and resized to 128x128 pixels. A sample slice from one such dataset is depicted in Figure 1a. The 'ground truth bounding boxes' for each slice were created by calculating the relative coordinates of the smallest box that contains the entirety of the left atrium. ...
Context 4
... 'ground truth bounding boxes' for each slice were created by calculating the relative coordinates of the smallest box that contains the entirety of the left atrium. This was done by utilizing the corresponding mask of the left atria for every slice in the dataset ( Figure 1b). is used as the input to the network. ...
Citations
... With more data, there will likely be an improvement in the accuracy of the network predictions. The next steps in this project include classifying the LAA from the automatically segmented [5] DICOM data and taking measurements of the LAA ostium as this an important parameter for sizing of various LAA occlusion devices. ...
Atrial fibrillation, a common cardiac arrhythmia, can lead to blood clots in the left atrial appendage (LAA) of the heart, increasing the risk of stroke. Understanding the LAA morphology can indicate the likelihood of a blood clot. Therefore, a classification convolutional neural network was implemented to predict the LAA morphology. Using 2D images of 3D models created from MRI scans of fixed human hearts and a pre-trained network, an 8.7% error rate was achieved. The network can be improved with more data or expanded to classify the LAA from the automatically segmented DICOM datasets and measure the LAA ostia.
