A team of NRC-IIT (National Research Council, Institute for Information Technology) competed in last year's i2b2 challenge on clinical text mining. The results, as they were announced at the i2b2 workshop in November, showed that the NRC-IIT team had produced the top ranking results for all three subtasks. This Tamale seminar gives a unique opportunity to hear from the team members how they developed their systems and which design decisions led to the successes.

The three subtasks of the 2010 i2b2 challenge for clinical text mining were: (1) extraction, from hospital progress reports and discharge summaries, of medical concepts of the types 'problem', 'treatment', and 'test'; (2) assessing for each 'problem' concept to what degree they were present for the patient; and (3) between pairs of concepts, determine their relationship. For all tasks, the NRC-IIT team built systems around machine-learning algorithms such that they could learn from a broad range of textual, semantic, and syntactic features. Development took place on a set of 350 annotated training documents and 827 unannotated training documents; the test was done on an additional 477 documents for which annotations had been made.

Note that with several team members presenting, this Tamale seminar is expected to be longer than usual: clear your calendars for an hour and a half presentation.