Radiotherapy treatment necessitates accurate tracking of the tumor in real-time, often during free-breathing. However, in lung cancer, the respiration entails a significant displacement of the tumor during radiation. This movement, if not well accounted for, can lead to an under-radiation of the tumor or damaging surrounding healthy regions. It is therefore paramount to be able to follow the displacement of the tumor over the entire respiratory cycle. In deep learning applications, it is important to have enough data to capture reliable and representative motion patterns. However, obtaining large amounts of dynamic images is known to be difficult, especially when there is a need to use manually annotated images. Consequently, even incomplete data are worth being utilized. In this work, we propose a model capable of predicting lungs deformations to predict missing phases in a 4D CT lungs dataset, based on probabilistic motion auto-encoders. The model uses the information from a reference 3D volume obtained at the beginning of treatment and a set of 2D surrogate images to predict the next 3D respiratory volumes. The proposed model was evaluated on a free-breathing 4DCT dataset of 165 patients treated for lung cancer. We achieve a mean performance of 81.70% structural similarity, a mean square error of 3.02% and a negative local cross correlation of 81.43% on a hold-out test set comprised of 34 patients. The proposed model can also be used to complete missing respiratory phases in datasets of 4DCT scans of lungs.
Respiratory motion models aim at improving the quality of free breathing image acquisition protocols and yield increased targeting accuracy during image guided interventions. Respiratory motion can deviate pre-defined targets and trajectories determined preoperatively during treatment procedures. In this context, motion models offer a mean to estimate spatio-temporal displacements of the organ and correct the target position in real time during an intervention. To construct a motion model, data of the entire organ of interest must be acquired. However, existing techniques for 3D dynamic imaging have poor spatial and temporal resolution. Therefore, to capture the organ’s temporal behavior, series of dynamic 2D slices covering the entire organ are typically acquired. Then, these slices are reordered retrospectively according to their motion phase within the respiratory cycle and stacked to form 3D dynamic volumes known as 4D images (3D + t). On the other hand, while numerous metrics were proposed to assess the spatial quality of the reordering, little attention has been paid to metrics that assess the coherent temporal behavior of the reconstructed dynamic volumes. This work proposes a method combining image-based matching approach with manifold alignment and compares it with two state of the art slice reordering methods. Methods were evaluated on a dataset of 7 volunteers using new metrics to assess the spatial quality and the temporal behavior, with the proposed method outperforming in terms of both spatial and temporal quality.
According to the World Health Organization, cardiovascular diseases are the leading cause of death worldwide, accounting for 17.3 million deaths per year, a number that is expected to grow to more than 23.6 million by 2030. Most cardiac pathologies involve the left ventricle; therefore, estimation of several functional parameters from a previous segmentation of this structure can be helpful in diagnosis. Manual delineation is a time consuming and tedious task that is also prone to high intra and inter-observer variability. Thus, there exists a need for automated cardiac segmentation method to help facilitate the diagnosis of cardiovascular diseases. In this work we propose a deep fully convolutional neural network architecture to address this issue and assess its performance. The model was trained end to end in a supervised learning stage from whole cardiac MRI images input and ground truth to make a per pixel classification. For its design, development and experimentation was used Caffe deep learning framework over an NVidia Quadro K4200 Graphics Processing Unit. The net architecture is: Conv64-ReLU (2x) – MaxPooling – Conv128-ReLU (2x) – MaxPooling – Conv256-ReLU (2x) – MaxPooling – Conv512-ReLu-Dropout (2x) – Conv2-ReLU – Deconv – Crop – Softmax. Training and testing processes were carried out using 5-fold cross validation with short axis cardiac magnetic resonance images from Sunnybrook Database. We obtained a Dice score of 0.92 and 0.90, Hausdorff distance of 4.48 and 5.43, Jaccard index of 0.97 and 0.97, sensitivity of 0.92 and 0.90 and specificity of 0.99 and 0.99, overall mean values with SGD and RMSProp, respectively.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.