In today’s era of digital transformation, the application of computer-aided analysis of medical images is not only a big data problem but also those of true cost-effective analytic methods(Sakr, 2014). The ongoing push to apply more advanced deep learning and convolutional neural networks in medical image analysis is a welcome development. Gibson et al. describe deep learning as a deeply nested composition of many simple functions, usually convolutional networks, parameterized by variables. The particular composition of convolutional networks of services sometimes called the architecture, defines a parametric function that is used to optimized an objective or loss function usually using some form of gradient descent methodologies (Gibson et al., 2018).
In consideration of the NiftyNet design architecture and the unique challenges associated with medical image analysis and its associated complexities because of domain application requirement and characteristics of the data itself. The authors considered the following key associated data constraints data availability, dimensionality, and size, formatting, and properties. Hence the design objectives of NifyNet an open source deep learning platform for medical image analysis was to and help accelerate more flexible and accurate outcomes and to provide a standard mechanism for disseminating research outputs for the community to use, adapt and build other representative learning applications.
However, medical image analysis requires a considerable effort in a wide range of tasks for different pre-clinical and clinical workflow. These tasks usually encompass image segmentation, classification, detection, registration, reconstruction, enhancement, model representation, and generation. In related literature, numerous approaches and solution application have been suggested and adopted, but NifyNet provides general application support for this via its application class workflow. The system can accept data from a noisy source with a built-in optional image conditioner, generator, optimizer, and discriminator subsystems. As an example, the NifyNet model zoo contains both untrained networks for image segmentation with a trained network for tasks like multiorgan abdominal CT segmentation, wnet for brain tumor segmentation and simulator GAN for generating ultrasound images(Gibson et al., 2018).
In conclusion, the NifyNet platform as an open source python based (simple installation via pip install niftynet) implementation supports data loading, data augmentation, network architecture, loss functions and evaluation metrics (accuracy performance turning) that well suited for the varied medical image analysis and computer-assisted clinical intervention. The system enables medical researchers to rapidly develop machine-learning solutions for medical image segmentation, regression, image generation, and other representative learning applications (Gibson et al., 2018).
Gibson, E., Li, W., Sudre, C., Fidon, L., Shakir, D. I., Wang, G., . . . Vercauteren, T., (2018). NiftyNet: a deep-learning platform for medical imaging. Comput Methods Programs Biomed, 158 , 113–122. doi:10.1016/j.cmpb.2018.01.025
Sakr, S., & Gaber, M., (2014). Large scale and big data: Processing and management . Boca Raton, FL: CRC Press.