Extracting Patient Data from Tables in Clinical Literature - Case Study on Extraction of BMI, Weight and Number of Patients

Abstract

Current biomedical text mining efforts are mostly focused on extracting information from the body of research articles. However, tables contain important information such as key characteristics of clinical trials. Here, we examine the feasibility of information extraction from tables. We focus on extracting data about clinical trial participants. We propose a rule-based method that decomposes tables into cell level structures and then extracts information from these structures. Our method performed with a F-measure of 83.3% for extraction of number of patients, 83.7% for extraction of patient's body mass index and 57.75% for patient's weight. These results are promising and show that information extraction from tables in biomedical literature is feasible.

Disentangling the Structure of Tables in Scientific Literature

“…We have already performed several information extraction experiments [15] and in the future we are planning to develop a general methodology for information extraction from tables in biomedical literature that uses the presented approach as its basis. Our methodology can be also used as a basis for semantic analysis and querying of tables.…”

DeepTable: a permutation invariant neural network for table orientation classification

“…Tables in our corpus were extracted from scientific articles published in the PubMed Central (PMC) Open Access subset 1 which currently consists of approximately 1.5 million full-text articles from several domains in the area of biomedicine and life sciences. Roughly 72% of these articles contain one or more tables (Milosevic et al 2016) which are rather easy to extract as they have explicit mark-up in PMC's XML format.…”

Genomics data integration for knowledge discovery using genome annotations from molecular databases and scientific literature

“…These genomic regions are called quantitative trait loci (QTLs). A QTL region can easily contain thousands of genes including those that negatively influence the trait of interest [107]. Therefore, detecting the causative gene for breeding is of major importance.…”

“…However, it is still challenging to predict a causal or candidate gene which is directly associated with the trait of interest. The size of a QTL region can vary enormously depending on the number of markers used and the genome size of the plant under investigation but easily can range from hundreds of kb to several Mb [143] and a single QTL region can contain very many genes [107]. One way to mine candidate genes from a QTL region could be done using the existing knowledge of structural and functional annotations of genes.…”