Disease Prediction by Machine Learning Over Big Data From Healthcare Communities
Download as PPTX, PDF4 likes4,408 views
The document discusses a novel convolutional neural network (CNN)-based multimodal disease risk prediction algorithm aimed at predicting cerebral infarction using both structured and unstructured data. The proposed CNN-multimodal disease risk prediction (CNN-MDRP) algorithm demonstrates an accuracy of 94.8%, significantly outperforming traditional methods that only use structured data. The study highlights the limitations of using structured data alone for complex diseases and advocates for a more integrated approach to leverage the power of big data analytics in healthcare.
Disease Prediction by Machine Learning Over Big Data From Healthcare Communities
- 1. PRESENTED BY MD. ZABIRUL ISLAM 1507110 MIN CHEN, (SENIOR MEMBER, IEEE) PUBLISHED IN: IEEE JOURNAL & MAGAZINE ( VOLUME: 5 , PAGES;8869-8879, APRIL 2017) Disease Prediction by Machine Learning Over Big Data From Healthcare Communities
- 2. OUTLINE Motivation Introduction Dataset Evaluation Methods Cnn-Based Unimodal Disease Risk Prediction (CNN-UDRP) Algorithm Experimental Results Conclusion Department of Computer Science and Engineering 1
- 3. Motivation In our modern society ,with the improvement of living standards, the incidence of chronic disease is increasing. According to a Chinese report in 2015, 86.6% of deaths are caused by chronic disease. So with the development of big data analytics technology , accurate analysis of medical data can give benefits early disease detection. Department of Computer Science and Engineering 2
- 4. Introduction Existing model uses structured data to predict the patients of either high risk or low risk. But for a complex disease, structured data is not a good way to describe the disease. We propose a new convolutional neural network (CNN)-based multimodal disease risk prediction algorithm using structured and unstructured data from hospital. In this paper, we mainly focus on the risk prediction of cerebral infarction. Department of Computer Science and Engineering 3
- 5. DATASET Structured data (S-data):the patient's basic information such as the patient's age, gender and life habits and laboratory data . Unstructured Text data (T-data): the patient's narration of his/her illness, the doctor's interrogation records and diagnosis. We use (S-data,T-data,S&T data) to predict whether the patient is at high-risk of cerebral infarction. Department of Computer Science and Engineering 4
- 6. CNN-BASED UNIMODAL DISEASE RISK PREDICTION (CNN-UDRP) ALGORITHM • CNN-UDRP: Only uses on the text data to predict high risk of cerebral infarction. • CNN-MDRP: Uses for structured and unstructured text data for prediction. • For the processing of medical text data, we utilize CNN-based unimodal disease risk prediction (CNN- UDRP) algorithm which can be divided into the following five steps. Department of Computer Science and Engineering 5
- 7. CONT. REPRESENTATION OF TEXT DATA: Each word in the medical text is represented in the form of vector i.e.Word Embedding in NLP. CONVOLUTION LAYER OF TEXT CNN: Every time we choose s words of each word vector in the text as the representation row vector. POOL LAYER OF TEXT CNN: Select the max value of the n elements of each row to which play key role in the text. FULL CONNECTION LAYER OF TEXT CNN CNN CLASSIFIER: Department of Computer Science and Engineering 6
- 8. EXPERIMENTAL RESULTS For S-data, we use traditional machine learning algorithms i.e.to predict the risk of cerebral infarction disease. The highest accuracy of DT is 63% , recall of NB is 0.80 than other For S-data, the NB classification is the best in experiment. Department of Computer Science and Engineering 8
- 9. CONT.. The number of iterations increasing, the training error rate of the CNN-UDRP (T-data) decreases test accuracy increases when the number of iterations are 70, the training process of CNN-MDRP (S&T-data) algorithm is already stable We extract 10, 20,….,120 features from text by using CNN. When the feature number of text is smaller than 30, the accuracy and recall of CNN-UDRP (T-data) and CNN-MDRP (S&T-data) algorithms are smaller. Department of Computer Science and Engineering 7
- 10. CONT. The accuracy of CNN-UDRP (T-data) is 0.9420 and the recall is 0.9808 The accuracy of CNN-MDRP (S&T-data) is 0.9480 and the recall is 0.99923 As seen for S-data the corresponding accuracy is low, which is roughly around 50%. We find that by combining these two data, the accuracy rate can reach 94.80% to better evaluate the risk of cerebral infarction disease. Department of Computer Science and Engineering 9
- 11. CONCLUSION • For some simple disease, e.g., hyperlipidemia, only a few features of structured data can get a good description of the disease, • But for a complex disease, only using features of structured data is not a good way to describe the disease. • In this paper, we propose (CNN-MDRP) algorithm using structured and unstructured data from hospital with 94.8% accuracy. Department of Computer Science and Engineering 10
- 12. REFERENCES P. Groves, B. Kayyali, D. Knott, and S. van Kuiken, The`Big Data'Revolution in Healthcare: Accelerating Value and Innovation. USA: Center for US Health System Reform Business Technology Ofce, 2016. M. Chen, S. Mao, and Y. Liu, ``Big data: A survey,'' Mobile Netw. Appl.,vol. 19, no. 2, pp. 171209, Apr. 2014. P. B. Jensen, L. J. Jensen, and S. Brunak, ``Mining electronic health records: Towards better research applications and clinical care,'' NatureRev. Genet., vol. 13, no. 6, pp. 395405, 2012. Department of Computer Science and Engineering 11
- 13. Department of Computer Science and Engineering 12
- 14. Department of Computer Science and Engineering 12