DeepPhe: A Natural Language Processing System for Extracting Cancer Phenotypes from Clinical Records

Abstract

Precise phenotype information is needed to understand the effects of genetic and epigenetic changes on tumor behavior and responsiveness. Extraction and representation of cancer phenotypes is currently mostly performed manually making it difficult to correlate phenotypic data to genomic data. In addition, genomic data is being produced at an increasingly faster pace, exacerbating the problem. The DeepPhe software enables automated extraction of detailed phenotype information from Electronic Medical Records of cancer patients. The system implements advanced Natural Language Processing and knowledge engineering methods within a flexible modular architecture, and was evaluated using a manually-annotated dataset of the University of Pittsburgh Medical Center (UPMC) breast cancer patients. The resulting platform provides critical and missing computational methods for computational phenotyping. Working in tandem with advanced analysis of high-throughput sequencing, these approaches will further accelerate the transition to precision cancer treatment.

A frame semantic overview of NLP-based information extraction for cancer-related EHR notes

“…Frame Elements References CANCER DIAGNOSIS NAME: cancer type [44], [45], [42], [46], [47], [48], [49], [50], [51] ANATOMICAL SITE: the location description of the finding (including primary and metastatic sites) [45], [52], [42], [53], [54], [55], [25], [27], [56], [57] HISTOLOGY: histological description (e.g. carcinoma) [44], [52], [58], [55], [53], [54], [4], [27], [59], [43], [57] GRADE: appearance of the cancerous cells, can be frame with further information (GRADING VALUE) [44], [52], [54], [4], [48], [27], [59], [60], [43], [61], [62] INVASION TYPE: the stage or level of invasion [52] TUMOR BLOCK: tissue cores removed from regions of interest in paraffinembedded tissues (e.g.…”

“…0.6 mm in diameter) [52] TISSUE BANK: identifiers about location of tissue samples within an institution [52] STATUS: whether confirmed, suspected and there is no evidence of finding (e.g. probable, definite, without) [42], [57] RECURRENT STATUS: the value of recurrent status [42], [63], [57] TEMPORAL INFORMATION: refers to information about time (e.g., year, month, and date, 2007-08-04) [42], [57] SPECIMEN TYPE: the type of specimen involved in diagnosis [53] LATERALITY: describes the side of a paired organ associated with origin of the primary cancer [53], [54], [25], [48], [64], [27], [51] TUMOR SIZE: how large across the tumor is at its widest point (part of cancer staging) [52], [53], [54], [25], [48], [65], [59], [60], [62], [57] TNM STAGE: cancer staging system, can be a separate frame with further information (TNM CLASSIFICATION) [55], [53], [2], [3], [25], [66], [67], [60], [50], [40], [61] EXTENSION: direct extension of tumor [53] UNCERTAINTY: used to differentiate clinical suspicions from conclusive findings (e.g., possible, likely) [68],…”

“…NEGATION: existence/negation of diagnosis (e.g., no, positive) [68], [57] DISEASE: disease related concepts related to diagnosis (disease stage and severity) [11], [25] DISEASE EXTENT: determine extent of disease (e.g., non-invasive, invasive, or metastatic) [68], [69], [57] STAGE: the overall stage of cancer (e.g. Stage 0, Stage I) [70], [52], [42], [54], [25], [69], [40], [43], [61], [57] EXISTENCE: existence description of the finding [45] TEMPORAL MODIFIER: temporal modifiers of the finding [45] ASSOCIATION: associations with other findings (may or may not be related to cancer) (e.g. causal, differential interpretation, and co-occurring) [45] BREAST CANCER DIAGNOSIS STATUS OF CANCER TYPES: presence or absence of various types of breast cancer (e.g.…”

“…Although a majority of the methods applying deep learning in cancer research are for processing images (e.g. mammograms) [16] [17] [18] [19] and gene expression profiles [20][21] [22], more recently, deep-learning based NLP systems are gaining prominence for cancer information extraction from EHRs [23][24] [25] [26] [27]. For example, Gao et al [26] implemented a hierarchical attention network for extracting some of the crucial clinical oncology data elements such as primary cancer site and histological grade which are gathered by cancer registries.…”

Longitudinal Visual Analytics for Unpacking the Cancer Journey

“…Existing cTAKES 12 pipelines were extended to extract cancer information, to use rules to infer higherlevel summaries, and to store results in a Neo4j graph database (www.neo4j.com). The initial DeepPhe architecture is described in detail by Savova, et al 7 .…”