With the background work done in the previous posts, let’s get to the meat of the problem. I’m currently working with a team from an excellent MIT course, Secondary Analysis of Health Records, led by Dr. Leo Anthony Celi. My team is composed of 4 clinicians and 3 …

View comments.

In the previous post I showed a convenient way to navigate the ICD9 hierarchy with Python, now let’s use that to extract the full taxonomy of ICD9 codes for patients who we’ll using to train a classifier. In this post we’ll be extracting the original ICD9 codes …

View comments.

ICD9, the International Statistical Classification of Diseases, is one of the common coding systems often used in medical databases for indicating patient conditions. A handy feature of it is that it’s a hierarchical system. Just as the Dewey Decimal System lets someone quickly find a book on geometry (library …

View comments.

One tricky thing to deal with in the transition from MIMIC II to MIMIC III has been that IDs have changed and are sometimes missing, so work that a collaborating team did to annotate patient notes in MIMIC II is not translatable to MIMIC III. This notebook shows one way …

View comments.

One of the hardest parts of learning Deep Learning is just figuring out how to start! A particularly annoying part of the process is getting access to a GPU environment - GPUs aren’t absolutely necessary for learning DL, but they are an essential part of making problems feasible to solve …

View comments.

Installing a container orchestration layer will allow us to quickly deploy anything in the huge ecosystem of available docker services. This can include database servers like Postgres or MariaDB, message brokers like RabbitMq, medical informatics servers like HAPI FHIR, or our own services. There are several options, including Kubernetes, Docker …

View comments.

We have several options for a data science front end, but my favorite is Jupyter Hub. Jupyter is the rebranded version of IPython, now generalized beyond Python to include R, Julia, Scala, and many other languages. Jupyter notebook provides a web-based interface that will allow us to create analytics experiments …

View comments.

Now that we have a secure private network and a bastion node setup, we can start configuring internal servers. Our goal is to configure as few servers as possible - not only is this easier (hooray for lazy sys admins!), it’s also more repeatable and more secure as it’s …

View comments.

I do most of my analysis now on cloud infrastructure as much as possible. The primary advantages are:

1. I can quickly scale compute resources as needed for large problems
2. I can access unlimited cheap storage (e.g. S3) as needed
3. My hardware doesn’t go obsolete - as processing, disks, and …

View comments.

As a final step we’ll configure Pelican to enable Jupyter notebook integration. This will let us quickly copy in analysis .ipynb files and include math, markdown, and figures. In further configuration we can even enable Pelican plugins for UML diagrams to add text-generated activity/state diagrams, or can add …

View comments.