AI Case Study
Researchers propose a new method for automating the contouring of high-risk clinical target volumes using neural networks
Researchers at The University of Texas MD Cancer Center have proposed a machine learning method using neural networks for automatically delineating clinical target volumes in patients. Clinical target volumes (CTV) are the areas containing tumours that need to be accurately defined prior to radiation therapy for cancer occurring in the head and neck. The research indicates good agreement with physicians' manual contouring while being quicker.
Healthcare Providers And Services
Science Daily: "The model uses the gross tumor volume and distance map information from surrounding anatomic structures as its inputs. It then classifies the data to identify voxels -- three-dimensional pixels -- that are part of the high-risk clinical target volumes. In oropharyngeal cancer cases, the head and neck are usually treated with different volumes for high, low and intermediate risk... In addition to potentially reducing inter-physician variability and allowing comparisons of outcomes in clinical trials, a tertiary advantage of the method is the speed and efficiency it offers. It takes a radiation oncologist two to four hours to determine clinical target volumes... Using the Maverick supercomputer at the Texas Advanced Computing Center (TACC), they were able to produce clinical target volumes in under a minute. Training the system took the longest amount of time, but for that step too, TACC resources helped speed up the research significantly.
The project is specifically intended to help low-and-middle income countries where expertise in contouring is rarer, although it is likely that the tools will also be useful in the U.S.... such a tool could also greatly benefit clinical trials by allowing one to more easily compare the outcomes of patients treated at two different institutions."
From the research paper: "When treating head and neck (H&N) cancer, radiation therapy prevails as the principal nonsurgical treatment op- tion. For this site in particular, the complexity of radiation treatment planning and the time required to delineate the target and normal tissue volumes are significantly increased (8) owing to the large number of organs at risk located near H&N tumors.
Manual delineation of clinical target volumes (CTVs) remains a time-consuming task in radiation oncology. CTVs are tissue volumes that contain the demonstrable gross tumor volume (GTV) and provide coverage for any suspected microscopic disease and pathways of tumor spread such as regional lymph nodes (1). Because the radiation dose is prescribed to these volumes and adequate coverage is required to achieve cure, accurate CTV delineation is essential in radiation therapy. Although established guidelines are available to delineate site-specific CTVs, these volumes are still subject to high intra- and interobserver variability for most treatment sites."
As stated in the research paper: "By implementing a DSC-based threshold selection function, our DNN auto-delineation algorithm accurately identified physician patterns to predict clinically acceptable high-risk CTV contours. Our models allowed for the prediction of new volumes within a few minutes and have the potential to greatly reduce physician contouring time. Most of the predicted high-risk CTVs were in close agreement with the physician manual contours and could be implemented clinically with only minor or no changes."
The researchers used "stacked auto-encoders owing to their ability to speed up training and provide improvement in predictions by initializing weights through unsupervised learning (17). During unsupervised learning, only the input data are provided, and the auto-encoders learn a general representation of the data set. Hidden layer neurons were activated using the logistic function. After this unsupervised learning step, we trained the output layer through supervised learning and used cross-validation to fine-tune the network architecture. During the supervised learning step, our algorithm fine-tuned the architecture by updating the network’s weights to match the training set’s inputs to the training set’s known output. Our deep auto-encoders are composed of 2 hidden layers, followed by a soft-max layer for binary classification." Note that this is actually semi-supervised learning because some unlabelled data to initialise weights for training a classification model with labelled data.
According to the research paper, "52 oropharyngeal cancer patients (11 base of tongue node-negative, 15 base of tongue node-positive, 15 tonsil node-negative, and 11 tonsil node-positive) who had under- gone curative-intent intensity modulated radiation therapy for H&N squamous cell carcinoma from January 2006 to August 2010 at The University of Texas MD Anderson Cancer Center were selected from an institutional review boardeapproved protocol. All patients had available simulation CT scans with previously manually contoured GTVs (primary and nodal, as applicable) and high-risk CTVs used for treatment planning."