Medical Imaging is an important tool utilized for diagnosing, monitoring and treating medical conditions. In medical imaging, anatomical or pathological structures may be difficult to distinguish, thus medical image segmentation is utilized to make those structures clearer. Manual segmentation approaches are tedious, and expensive. Deep convolutional neural networks have shown to work well at segmentation, however, they require large sets of labeled training data, which is difficult to come by.
Researchers at Arizona State University have developed a novel interactive training strategy for medical image segmentation. This strategy interactively refines the segmentation map through several iterations of training for continuous improvement and prediction. A convolutional neural network is trained with user simulated inputs to edit the segmentation and improve segmentation accuracy. When tested on different datasets, this strategy showed superior performance in comparison to other strategies.
This semi-automatic strategy provides user interactions to the network as additional input or feedback to guide the neural network, correct segmentation error, improve segmentation accuracy and provide a feedback control loop.
Finally, using interactive network on top of the state-of-the-art segmentation architecture, improves the prediction accuracy further, compared to when the base model is a simple
encoder decoder architecture.
Benefits and Advantages
Speeds up annotations
Corrects segmentation error and improves segmentation accuracy
Provides maximum performance boost in just two to three user-inputs
Can refine existing methods
Is able to better predict boundaries of an object
Continuously improves with each interaction from new information provided by the user and updated predictions
Generalized and works on a variety of types of images
Performance demonstrated on a prostate dataset and a heart, spleen, pancreas and hippocampus dataset
For more information about this opportunity, please see
For more information about the inventor(s) and their research, please see