Abstract
Deep learning has achieved remarkable performance in many classification tasks such as image processing and computer vision. Due to its impressive performance, deep learning techniques have found their way into natural language processing tasks as well. Deep learning methods are based on neural network architectures such as CNN (Convolutional Neural Networks) with many layers. Deep learning methods have shown state of-the-art performance on many classification tasks through several research works. It has shown great promise in many NLP (Natural language processing) tasks such as learning text representations. In this paper, we study the possibility of using deep learning methods and techniques in clinical documents classification. We review various deep learning-based techniques and their applications in classifying clinical documents. Further, we identify research challenges and describe our proposed convolutional neural network with residual connections and range normalization. Our proposed model automatically learns and classifies clinical sentences into multi-faceted clinical classes, which can help physicians to navigate patients' medical histories easily. Our propose technique uses sentence embedding and Convolutional Neural Network with residual connections and range normalization. To the best of our knowledge, this is the first time that sentence embedding and deep convolutional neural networks with residual connections and range normalization have been simultaneously applied to text processing. Lastly, this work follows a generalized conclusion on clinical documents classification and references.
Highlights
Unstructured narrative text such as clinical notes, pathology notes, radiology reports, discharge summaries, etc. are important documents stored in Electronic Health Records (EHRs) [3] and contains valuable information [3] which can be useful for patient care [4] and clinical research [5], [6] such as extraction of relations from clinical notes [7]
A lot of research work has shown that most electronic health records are documented using narrative text [46], [47] and it is always preferred by clinicians as it is the most natural and expressive way for documenting the clinical patients encounters
In the last few years, a combination of deep neural networks and its derivatives (CNN and recurrent neural network (RNN)) and distributional word representations have been preferred for the clinical documents classifications for the following reasons: i) The ability of word embeddings to capture the semantic relationships of words[91] ii) The ability of deep learning methods to make use of available data to extract important features without requiring a lot of labeled data (NLP) [92]
Summary
Information technology is playing an increasingly important role in medical practice. A comprehensive patient’s clinical record contributes to quality and efficient healthcare of a patient during hospitalization and during subsequent follow-up visits, as they can provide a complete and accurate chronology of treatments, patient results and future plans of care [9], [10] Despite their utility, the unstructured nature of clinical documents makes it challenging to retrieve and extract relevant information about patients during care episodes or retrieve relevant information about a population during clinical research. Text classification of clinical documents would pave the way to categorize the medical sentences into multi-faceted categories of information This would allow physicians to analyze, summarize, and explore content of medical documents related to various patients’ diseases and conditions
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
