Clinical Discharge Summaries contain a huge amount of data but are difficult to obtain and to be used in analysis because of its unstructured narrative format. Also, they contain spelling mistakes, abbreviations which makes it difficult to summarize. With vast amount of information, there is a need to have a proper segregation of important details like Symptoms, History, Medications to have a better and efficient classification.
To save the overall time and energy of the patient and increase the effectiveness of the physician in saving the lives of the sufferers, we suggest a classification of the Clinical Discharge Summaries text, which aims to automatically predict the diagnoses needed for a patient based on clinical notes. The reminiscence is represented by a raw text file with doctor's entries about the patient including his / her age, problems described in normal flow, history of the patient and so on.
We are also investigating the automated classification of patient discharge notes into standard disease labels which includes the name of the diagnostic procedure required. In this approach, we use Convolutional Neural Networks to classify and represent complex features from the medical discharge summaries using the MT sample dataset. We make use of GloVE to have a pretrained model learn from it.
We implemented our CNN model on MT samples dataset with multiple layers being added and removed to improve the efficiency. Our training accuracy increases with every epoch and reaches 0.87 and the validation accuracy is at 0.89 as shown in the graph 1. The model during its validation set also achieved a recall score of 0.888, precision of 0.98 and an F1 score of 0.939.