[01] Jabusile Madondoro, Tuberculosis Laboratory, Mpilo National and Referral Hospital, Bulawayo, Zimbabwe. [02] Kernan Mzelikahle, Computer Science Department, National University of Science and Technology, Bulawayo, Zimbabwe. [03] Dumisani John Hlatywayo, Applied Physics Department, National University of Science and Technology, Bulawayo, Zimbabwe. [04] John Trimble, Industrial Engineering Department, Tshwane University of Technology, Tshwane, South Africa.

Tuberculosis infections that occur both within the medical facility environments and general population have long been attributed to unrecognised strains of the bacteria and to previously unsuccessful treatment. Identifying active pulmonary TB, for both the initial activation of the disease and recurring disease, is very crucial in breaking the transmission cycle of the disease, particularly in low resourced countries of the developing world. In this paper, a Long Short-Term Memory (LSTM) network was adopted for use by training it using radiological data and other admission specific data into a medical facility. This data is usually made available upon a patient’s presentation onto a medical facility by the patient themselves, or through access to historical data. The objective of the LSTM network in this study is to complement the physician’s expert opinion on point of presentation of the patient into a medical facility. This study was set up as a non-concurrent prospective study, using data from the National Tuberculosis Laboratory at Mpilo Hospital in Bulawayo, Zimbabwe. Participants were identified through access to laboratory historical data, and the participates were divided into two groups. The first group is referred to as a derivation group and had a total of 5630 isolated instances of suspected active pulmonary TB. The second group was identified as the validation group and had a total of 1388 isolated instances of suspected active pulmonary TB as was determined at the point of presentation. The Long Short-Term Memory (LSTM) network was adopted and employed to predict active recurrent TB cases given the data available on point of presentation. The results of the LSTM prediction were contrasted with both the physicians’ assessments and results of subsequent investigations. The accuracy of both the physicians’ assessments and LSTM predictions were measured by calculating a c-index based on the area under the receiver operating characteristics curve. The results of this process indicate that the LSTM network significantly outperformed the physicians’ assessments, with calculated c-indices of 0.947 ± 0.028 and 0.61 ± 0.045, respectively (p < 0.05). By applying the LSTM network to the validation group, similar results are obtained where the corresponding c-indices were 0.923 ± 0.056 and 0.716 ± 0.095, respectively. In conclusion, the LSTM network was shown to have higher potential in identifying patients with recurring pulmonary TB, more accurately than physicians’ clinical assessment. This property may prove useful in low resourced countries where health facilities have very high doctor-patient-ratios.

Artificial Neural Networks, Tuberculosis, Active Tuberculosis, Recurrent Tuberculosis, Long Short-Term Memory Network

[01] El-Solh, A. A., Hsiao, C. B., Goodnough, S., Serghani, J. and Grant, B. J., 1999. Predicting active pulmonary tuberculosis using an artificial neural network. Chest, 116 (4), pp. 968-973. [02] Er, O., Temurtas, F. and Tanrıkulu, A. Ç., 2010. Tuberculosis disease diagnosis using artificial neural networks. Journal of medical systems, 34 (3), pp. 299-302. [03] Elveren, E. and Yumuşak, N., 2011. Tuberculosis disease diagnosis using artificial neural network trained with genetic algorithm. Journal of medical systems, 35 (3), pp. 329-332. [04] Pavlou, A. K., Magan, N., Jones, J. M., Brown, J., Klatser, P. and Turner, A. P., 2004. Detection of Mycobacterium tuberculosis (TB) in vitro and in situ using an electronic nose in combination with a neural network system. Biosensors and Bioelectronics, 20 (3), pp. 538-544. [05] Khaliq, A., Batool, S. A. and Chaudhry, M. N. Seasonality and trend analysis of tuberculosis in Lahore, Pakistan from 2006 to 2013. J. Epidemiol. Global Health 2015, 5, 397-403. [06] Willis, M. D., Winston, C. A., Heilig, C. M., Cain, K. P., Walter, N. D. and Mac Kenzie, W. R. Seasonality of tuberculosis in the United States, 1993 - 2008. Clin. Infect. Dis. 2012, 54, 1553-1560. [07] Naranbat, N.; Nymadawa, P.; Schopfer, K.; Rieder, H. L. Seasonality of tuberculosis in an Eastern-Asian country with an extreme continental climate. Eur. Respir. J. 2009, 34, 921-925. [08] Soetens, L. C., Boshuizen, H. C.; Korthals Altes, H. Contribution of seasonality in transmission of mycobacterium tuberculosis to seasonality in tuberculosis disease: A simulation study. Am. J. Epidemiol. 2013, 178, 1281-1288. [09] Karim, S., Churchyard, G., Karim, Q. and Lawn, S. (2009). HIV infection and Tuberculosis in South Africa: An urgent need to escalate the public health response. Lancet, 374, 921-933. [10] Claassens, M. M.; Du Toit, E.; Dunbar, R.; Lombard, C.; Enarson, D. A.; Beyers, N.; Borgdorff, M. W. (2013) Tuberculosis patients in primary care do not start treatment. What role do health system delays play? Int. J. Tuberc. Lung Dis, 17, 603-607. [11] Karim, S. S. A., Naidoo, K., Grobler, A., Padayatchi, N., Baxter, C., Gray, A., Gengiah, T., Nair, G., Bamber, S., Singh, A., et al. Timing of antiretroviral drugs during tuberculosis therapy. N. Engl. J. Med. 2010, 362, 697-706. [12] Narula, P., Sihota, P., Azad, S., and Lio, P. Analyzing seasonality of tuberculosis across Indian states and union territories. J. Epidemiol. Global Health 2015, 5, 337-346. [13] Saha, S. and Raghava, G. P. S., 2006. Prediction of continuous B-cell epitopes in an antigen using recurrent neural network. Proteins: Structure, Function, and Bioinformatics, 65 (1), pp. 40-48. [14] Lee, M. J. and Chen, J. T., 1993. Fluid property predictions with the aid of neural networks. Industrial & engineering chemistry research, 32 (5), pp. 995-997. [15] Fojnica, A., Osmanović, A. and Badnjević, A., 2016, June. Dynamical model of tuberculosis-multiple strain prediction based on artificial neural network. In 2016 5th Mediterranean Conference on Embedded Computing (MECO) pp. 290-293, IEEE. [16] Yang, H., Tang, H., Chen, X. X., Zhang, C. J., Zhu, P. P., Ding, H., Chen, W. and Lin, H., 2016. Identification of secretory proteins in mycobacterium tuberculosis using pseudo amino acid composition. BioMed research international, 2016. [17] Lakhani, P. and Sundaram, B., 2017. Deep learning at chest radiography: automated classification of pulmonary tuberculosis by using convolutional neural networks. Radiology, 284 (2), pp. 574-582. [18] Wang, K. W., Deng, C., Li, J. P., Zhang, Y. Y., Li, X. Y. and Wu, M. C., 2017. Hybrid methodology for tuberculosis incidence time-series forecasting based on ARIMA and a NAR neural network. Epidemiology & Infection, 145 (6), pp. 1118-1129. [19] Mzelikahle, K., Mapuma, D. J., Hlatywayo, D. J. and Trimble, J., (2017). Optimisation of Self Organising Maps Using the Bat Algorithm. American Journal of Information Science and Computer Engineering, 3 (6), pp. 77-83. [20] Mzelikahle, K., Trimble, J. and Hlatywayo, D. J., (2018). A Hybrid Technique Between BOSOM and LSTM for Data Analysis. International Journal of Mathematics and Computational Science, 4 (4), pp. 128-138. [21] Yan, W., Xu, Y., Yang, X.; Zhou, Y. (2010) A hybrid model for short-term bacillary dysentery prediction in Yichang City, China. Jpn. J. Infect. Dis., 63, 264-270. [22] Wei, W., Jiang, J., Liang, H., Gao, L., Liang, B., Huang, J., Zang, N., Liao, Y., Yu, J., Lai, J. and Qin, F., 2016. Application of a combined model with autoregressive integrated moving average (ARIMA) and generalized regression neural network (GRNN) in forecasting hepatitis incidence in Heng County, China. PloS one, 11 (6), p. e0156768. [23] Zheng, Y. L., Zhang, L. P., Zhang, X. L., Wang, K. and Zheng, Y. J., 2015. Forecast model analysis for the morbidity of tuberculosis in Xinjiang, China. PloS one, 10 (3), p. e0116832. [24] DeVries, P. M., Thompson, T. B. and Meade, B. J., 2017. Enabling large-scale viscoelastic calculations via neural network acceleration. Geophysical Research Letters, 44 (6), pp. 2662-2669. [25] Goh, G. B., Siegel, C., Vishnu, A., Hodas, N. O. and Baker, N., 2017. Chemception: A deep neural network with minimal chemistry knowledge matches the performance of expert-developed qsar/qspr models. arXiv preprint arXiv: 1706.06689. [26] Fei, Y., Hu, J., Gao, K., Tu, J., Li, W. Q. and Wang, W., 2017. Predicting risk for portal vein thrombosis in acute pancreatitis patients: A comparison of radical basis function artificial neural network and logistic regression models. Journal of critical care, 39, pp. 115-123. [27] Azeez, A., Obaromi, D., Odeyemi, A., Ndege, J. and Muntabayi, R., 2016. Seasonality and trend forecasting of tuberculosis prevalence data in Eastern Cape, South Africa, using a hybrid model. International journal of environmental research and public health, 13 (8), p. 757. [28] Glaziou, P., Sismanidis, C., Floyd, K. and Raviglione, M. (2015) Global Epidemiology of Tuberculosis. Cold Spring Harb. Perspect. Med., 5, 1-18. [29] Corbett, E. L., Marston, B., Churchyard, G. J., and De Cock, K. M. (2006) Tuberculosis in sub-Saharan Africa: Opportunities, challenges, and change in the era of antiretroviral treatment. Lancet, 367, 926-937.