MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Deep Learning with Medical Images

C Parmar*, R Zeleznik*, Dana-Farber Cancer Institute, Boston, MA



Presentations

(Thursday, 8/2/2018) 10:00 AM - 12:00 PM

Room: Davidson Ballroom B

Artificial Intelligence (AI) algorithms, deep learning in particular, have demonstrated remarkable progress in image recognition tasks. Radiographic images have different data structure as compared to the regular natural images. Moreover unlike natural images, medical images provide the complete 3D volumetric information of an object. Therefore, it is required to identify the most suitable AI methods for radiological data and standardize their applications. In this session we will discuss different image preprocessing methods for processing volumetric medical imaging data and convert them to suitable data structures for the development of AI applications. We will also discuss different deep learning architectures related to different clinical tasks like detection, segmentation, characterization and monitoring.

In the lab session we will apply image segmentation to CT scans, an application of deep learning to medical imaging. Image segmentation is particularly important as it often represents the basis of processing pipelines through identification and labeling of the region of interest for further processing. We will focus on the U-Net architecture, a powerful and state of the art segmentation network that performs very well on smaller data sets too. The lab session will be held using Jupyter Notebooks, Python and Keras with Tensorflow. Attendees will be introduced to U-Net and have the opportunity to test different network settings on medical images. This lecture will provide an overview of the usage of Keras with Tensorflow. Furthermore, attendees will test different network architectures and assess their effects on the segmentation performance.

Learning Objectives:
1. Image processing and data normalization methods for developing deep learning applications of medical image data.
2. Formatting medical image data for pre-trained CNNs of natural images.
3. Basic understanding of the neural network API Keras and the deep learning framework Tensorflow.
4. Application and evaluation of different network layers and different network structures for image segmentation.

Handouts

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email