dbeanm / hiutils

HIUTILS

A collection of utilities for health informatics

This is pre-alpha, anything might change, i.e. not ready for production use.

Application areas

Text annotation / NLP

Ontologies

Knowledge graphs

Statistics / summary data

Installation

pip install hiutils

Annotations

Overview

We assume that annotations are in the format:

{
	document_id: {
		entities: {
			entitiy_id: {
				...properties...,
				cui : "concept_id",
				meta_anns: {
					'meta_ann_name': {'value': 'meta_ann_value',
					'confidence': confidence,
					'name': 'meta_ann_name'},
					...other meta...
				}
			}
		}

	}
}

Basic process

The aim is to:

1. keep only some annotations based on context
2. convert from document->concepts to patient->concepts
3. limit to a subset of concepts relevant to our project
4. group some specific concepts into more general concepts e.g. specific subtypes of a disease -> any occurence of a that disease

To achieve these aims:

• 1 filter by meta_anns:

filtered = hi.annotations.filter_anns_meta(anns, {'Subject': ['Other']}, inplace=False)

• 2 aggregate to patient level

agg = hi.annotations.aggregate_docs(filtered, item2doc=pt2doc)

• 3+4 group relevant concepts and drop other concepts

groups = {'Group 1': set(['286933003', '70582006']), 'My other group': set(['60046008'])}
merged = hi.merge_concepts(agg, groups, keep_empty=False)