Clinical Text Classification with Rule-based Features and Knowledge-guided Convolutional Neural Networks

Abstract

Clinical text classification is an important problem in medical natural language processing. Existing studies have conventionally focused on rules or knowledge sources-based feature engineering, but only a few have exploited effective feature learning capability of deep learning methods. In this study, we propose a novel approach which combines rule-based features and knowledge-guided deep learning techniques for effective disease classification. Critical Steps of our method include identifying trigger phrases, predicting classes with very few examples using trigger phrases and training a convolutional neural network with word embeddings and Unified Medical Language System (UMLS) entity embeddings. We evaluated our method on the 2008 Integrating Informatics with Biology and the Bedside (i2b2) obesity challenge. The results show that our method outperforms the state of the art methods.

Early Prediction of Acute Kidney Injury in Critical Care Setting Using Clinical Notes

“…We also explored deep learning models for AKI prediction. We used Knowledge-guided Convolutional Neural Networks (CNN) to combine word features and UMLS CUI features [14]. It used pre-trained word embeddings and CUIs embeddings of clinical notes as the input.…”

“…It used pre-trained word embeddings and CUIs embeddings of clinical notes as the input. We adopted the same architecture and parameter settings for Knowledge-guided CNN as in [14]. To address imbalance, We experimented random under-sampling with the following training class ratio, 1:1, 1:2 and 1:3.…”

Data‐driven discovery of probable Alzheimer's disease and related dementia subphenotypes using electronic health records

“…T A B L E 6 Variables with P value > .05 (Chi-square test) between subphenotypes P value ≤.05.unstructured clinical text from EHRs, including clinician encounter notes, which may contain additional information about an individual's comorbidities and treatment. In the future, we will apply our extensive work in natural language processing to analyze such data 28,29. 5 | CONCLUSION Using routinely collected longitudinal EHR data and ML algorithms, we computationally derived probable AD and related dementia subphenotypes that can potentially guide improved diagnosis and treatment of AD patients.…”

Clinical Text Mining of Electronic Health Records to Classify Leprosy Patients Cases

“…A rule-based model was compared with other machine learning models like fph, logistic regression, and decision tree [5]. In this, a rule based feature classification along a deep learning technique was studied for effective disease classification [3].Text mining and data processing of EMR patients using named entity recognition, data cleansing, data transformation,reduction and integration [1]. Using rule classification and logistic regression are hybrid approach models of both the algorithms in applied and compared through classification parameters [4].…”