Technical Description

Technical Description

Introduction

MedLEE has four main components, as represented below

1. The Preprocessor:

Words are matched to the semantic lexicon and coded with their semantic type

Irrelevant 'stop words' and known irrelevant phrases are removed from the text.

2. The parser

MedLEE has a set of 'sentence grammars'. These are common patterns of semantic types which appear is radiology texts.

The grammars are predominantly semantic, though some syntactic information is used.

If the sentence matches a grammar rule, a frame for the information is generated. If the sentence does not parse, MedLEE will try and reparse smaller segments of the sentence.

3. The Phrase Regularizer

Uses a series of mapping rules to resolve split patterns. Split patterns happen when multi-word phrases are discontinuous in the text.

4. The Encoder

Identified semantic units are matched to the Medical Entities Dictionary (MED), a controlled vocabulary in the CIS.

The MED contains preferred terms for concepts and is used to reduce redundancy and ambiguity in the case of synonymy.