Time Series Analysis with LSTM Based Deep Learning Network in Neonatal Laboratory Findings

Abstract

Research Problem/Questions – Laboratory findings and vital values should be monitored regularly in babies hospitalized in neonatal intensive care. Infant blood vessels are quite weak compared to a normal person. At the same time, due to the very low blood volume, blood analyses cannot be performed continuously or all the desired laboratory findings cannot be obtained with insufficient blood. Especially in premature babies (preterm) who stay in neonatal intensive care and are disadvantaged compared to other babies, these problems are experienced more frequently and in addition to these, it is known that preterm morbidity is much higher. Short Literature Review – Time series analyses are used in many fields such as finance, climate, meteorology, medical and military. Time series analysis, which started to develop with methods that produce solutions for linear data, now uses advanced methods such as deep learning architectures, transformer architectures and machine learning. In the medical field, time series analysis has many different uses. In this study, a time series analysis that makes forward-looking predictions by evaluating laboratory findings obtained from infants over a certain period of time has been performed. Deep learning model based on LSTM network architecture is used as time series analysis method. Methodology – In this study, 161 data were obtained from 22 healthy babies with sepsis, rds, nec, ikk, diaphragmatic, pneumothorax diseases and 22 healthy babies in the neonatal intensive care unit and the laboratory findings obtained within a certain period of time for each baby were converted into time series data. In the time series group created for each of 22 babies, minimum 5 and maximum 15 laboratory findings were obtained. CRP, hemoglobin and bilirubin values, which are frequently monitored, were selected as laboratory findings. The LSTM model was trained with the time series data created for each baby. Results and Conclusions – When the model evaluation results were analyzed, it was observed that the accuracy of CRP data was very low. Since CRP values are a parameter that increases when infection increases in the body, it is a parameter that is not frequently monitored especially in healthy or non-infected infants. Therefore, CRP was the parameter with the least up-to-date data during this process. When these evaluations were examined, the parameter estimx"ated with the highest accuracy was hemoglobin with an accuracy of 43.63% and a mean squared error of 7.37. CRP parameter showed the lowest performance with 2274.9% mean squared error and 29.09% accuracy. Bilirubin remained at 36.36% accuracy level. Data acquisition and model development processes are ongoing and these initial results are promising for future studies.

Keywords

• Time Series Analysis
• LSTM
• Deep Learning
• Data Prediction
• CRP
• Hemoglobin
• Bilirubin
• Premature morbidity

Yenidoğan Laboratuvar Bulgularında LSTM Tabanlı Derin Öğrenme Ağı ile Zaman Serisi Analizi

Abstract

Yenidoğan yoğun bakımında kalan bebeklerde laboratuvar bulguları ve hayati değerler düzenli olarak takip edilmeli ve değerlendirilmelidir. Bebek kan damarları normal bir insana göre oldukça zayıf ve incedir. Aynı zamanda kan hacminin çok düşük olması sebebi ile sürekli olarak kan tahlili yapılamamakta veya alınan kan yetersizliği ile istenen tüm laboratuvar bulguları elde edilememektedir. Özellikle yenidoğan yoğun bakımında kalan ve diğer bebeklere göre dezavantajlı bulunan prematüre bebeklerde (preterm) bu olumsuzluklar daha sık yaşanmakta ve bunlara ek olarak preterm morbiditesinin çok daha yüksek olduğu da bilinmektedir. Bu çalışmada bebeklerden belirli bir zaman içerisinde elde edilen laboratuvar bulgularını değerlendirerek ileriye yönelik tahminler yapan bir zaman serisi analizi gerçekleştirilmiştir. Zaman seri analizi yöntemi olarak LSTM ağ mimarisine dayalı derin öğrenme modeli kullanılmıştır. Bu çalışma için 22 adet bebekten 161 veri elde edilmiş ve her bir bebek için belirli bir zaman içerisinde alınan laboratuvar bulguları zaman serisi verileri haline getirilmiştir. Laboratuvar bulguları olarak sıklıkla takip edilen CRP, hemoglobin ve bilirubin değerleri seçilmiştir. Her bebek için oluşturulan zaman seri verileri ile LSTM modeli eğitilmiştir. LSTM modelinin sonuçları incelendiğinde CRP değerinin tahmininde doğruluk değerinin %29.09’da kaldığı, en yüksek tahmin sonucunun ise %43.63 ile hemoglobin değerlerinde elde edildiği gözlemlenmiştir. Bilirubin değerleri için doğruluk oranı ise %36.36’dır. Kısıtlı veri seti ile elde edilen bu sonuçların umut vaat ettiği ve gelecek çalışmalar için önemli olduğu değerlendirilmiştir.

Keywords

• Zaman Seri Analizi
• LSTM
• Derin Öğrenme
• Veri Tahmini
• CRP
• Hemoglobin
• Bilirubin
• Prematüre Morbiditesi

References

1. [1] N. I. Sapankevych and R. Sankar, "Time Series Prediction Using Support Vector Machines: A Survey", IEEE Computational Intelligence Magazine, vol. 4, no. 2, pp. 24-38, 2009
2. [2] O.B. Sezer, M.U. Gudelek, A. M. Ozbayoglu, “Financial time series forecasting with deep learning: A systematic literature review: 2005–2019”, Applied Soft Computing, vol. 90, pp. 106-181, 2020
3. [3] T. Dimri, S. Ahmad, M. Sharif “Time series analysis of climate variables using seasonal ARIMA approach” J Earth Syst Sci, pp. 129-149, 2020
4. [4] R. B. Penfold, F. Zhang, “Use of Interrupted Time Series Analysis in Evaluating Health Care Quality Improvements”, Academic Pediatrics, vol. 13, pp.38-44, 2013
5. [5] S. S. Vakhare, R. R. Manza, M. M. Mhaske, “Time Series Analysis and Forecasting of Temperatures Records of Aurangabad District in Maharashtra”, International Journal for Modern Trends in Science and Technology, vol.6, pp. 291-295, 2020
6. [6] G. Box, G. Jenkins, G. Reinsel, G. Ljung, “Time Series Analysis: Forecasting and Control”, New Jersey, John Wiley & Sons,
7. [7] S. Makrıdakıs, M. Hibon, “ARMA Models and the Box–Jenkins Methodology”, Journal of Forecasting, vol.16, pp.147-163
8. [8] N.K. Ahmed, A.F. Atiya, N. El Gayar, H. El-Shishiny, “An Empirical Comparison of Machine Learning Models for Time Series Forecasting”, Econometric Reviews, vol. 29, pp.594-621, 2010

9. [9] B. Lim, S. Zohrem, “Time-series forecasting with deep learning: a survey”, Royal Society, vol. 379, Issue 2194, 2021
10. [10] S. Ahmed, I.E. Nielsen, A. Tripathi, S. Siddique, R.P. Ramachandran, G. Rassol, “Transformers in Time-Series Analysis: A Tutorial” Circuits Syst Signal Process, vol. 42, pp. 7433–7466, (2023)
11. [11] A. Kumar Dubey, A. Kumar, V. García-Díaz, A. Kumar Sharma, K. Kanhaiya, “Study and analysis of SARIMA and LSTM in forecasting time series data”, Sustainable Energy Technologies and Assessments, vol. 47, pp. 101474, 2021
12. [12] S. Siami-Namini, N. Tavakoli and A. Siami Namin, "A Comparison of ARIMA and LSTM in Forecasting Time Series”, 17th IEEE International Conference on Machine Learning and Applications (ICMLA), Orlando, FL, USA, pp. 1394-1401, 2018
13. [13] V.K.R. Chimmula, L. Zhang, “Time series forecasting of COVID-19 transmission in Canada using LSTM networks”, Chaos, Solitons & Fractals, vol. 135, pp. 109864, 2020
14. [14] Anonim, 2023, Preterm Birth [online], World Health Organization, https://www.who.int/news-room/fact-sheets/detail/preterm-birth [Ziyaret Tarihi: 15 Ekim 2023]
15. [15] Anonim, 2021, Dünya Prematüre Günü [online], T.C. Sağlık Bakanlığı, Halk Sağlığı Genel Müdürlüğü, https://hsgm.saglik.gov.tr/tr/haberler-cocukergen/dunya-premature-gunu.html , [Ziyaret Tarihi: 16 Ekim 2023]
16. [16] Anonim, 2023, Ölüm ve Ölüm Nedeni İstatistikleri, 2021 [Online], https://data.tuik.gov.tr/Bulten/Index?p=Olum-ve-Olum-Nedeni-Istatistikleri-2021-45715 , [Ziyaret Tarihi: 5 Mayıs 2023]
17. [17] O. Kapellou, “Blood sampling in infants (reducing pain and morbidity)”, BMJ Clinical Evidence, vol. 2009 0313, 2009
18. [18] S. Hochreiter, J. Schmidhuber, "Long Short-Term Memory", Neural Computation, vol. 9, no. 8, pp. 1735-1780, 1997