Large-scale medical equipment, extensively implemented in medical services, is of vital importance for diagnosis but vulnerable to various failures. Most of hospitals conduct regular maintenance for their medical equipment and have been suffering from medical-equipment-related incidents for years. Currently, the Internet of Medical Things (IoMT) has emerged as a crucial technology to monitor the real-time status of medical equipment. In this paper, we develop a IoMT system of Computed Tomography (CT) equipment in the West China Hospital of Sichuan University and meanwhile, select machine-learning algorithms to effectively store, preprocess and analyze the data. Specifically, a data-driven framework is proposed to predict the anomalies of CT equipment successfully. In this framework, the first step is data preprocessing, where we average the original non-uniform data and use linear interpolation to handle missing data. The second step is feature construction, where sliding time window is applied to fully reflect historical information. The third step is features selection, where the seven features that perform best are selected. Finally, we use two methods to split the training and test set, and apply random oversampling to deal with those imbalanced data before putting them into models. The results show that the prediction precision and recall of our method are 70% and 81%, respectively. The proposed method could distinguish the state of CT equipment and be used as a reference for practical maintenance by early prediction, where unexpected anomalies of medical equipment could be reduced. It also brings new insights about how to handle non-uniform and unbalanced time series data in the practical cases.
Research Article
Anomaly Prediction of CT Equipment Based on IoMT Data
https://doi.org/10.21203/rs.3.rs-1548704/v1
This work is licensed under a CC BY 4.0 License
Version 1
posted
You are reading this latest preprint version
anomaly prediction
CT equipment
Internet of Medical Things
multivariate time series classification
maintenance strategy
Currently, various medical equipment has been extensively implemented in all aspects of medical services, including disease diagnosis, patient condition monitoring and rehabilitation. Particularly, the large-scale digital radiology equipment such as Computed Tomography (CT), allowing for clear cross-sectional images of internal organs through X-rays, is of vital importance for medical facilities to treat patients. However, the CT equipment, embedding sophisticated operating systems, is vulnerable to various types of damages during the operation. Anomalies such as abnormal parameters, unstable current, damages of components and system outage, which occur unexpectedly during the equipment operation, have long plagued the hospitals as a problem. The equipment anomalies could result in unexpected delays in patient care, costly maintenance service and even serious patient incidents. In fact, the survey conducted by The Joint Commission (TJC) [1] shows that the “sentinel events” (i.e., safety accidents such as premature deaths, preventable diseases, severe injuries and disability accidents, the occurrence of which are not primarily related to patients’ illness [2]) associated with the failure of medical equipment is usually among the top ten reasons of medical accidents each year. It is reported that there were a total of 176 medical equipment-related incidents in the US, accounting for 2.9% of the total number of 6093 activities collected from 8 hospitals during the period 2004–2011 [3]. Therefore, medical facilities including hospitals and healthcare organizations must ensure high-level reliability of medical equipment to guarantee patients’ safety and meet the required operation standards.
To date, the maintenance strategies including Corrective Maintenance (CM) [4], Preventive Maintenance (PM) [5–10] and Predictive Maintenance (PdM) [11–14], etc. have been successfully applied to various fields such as mechanical engineering [15, 16], nuclear engineering [17], management science [18, 19] etc., which greatly improved the management level of those systems. However, the applications of maintenance models have not been thoroughly addressed on the medical equipment. Generally, most medical facilities perform their equipment maintenance by merely following the manufacturer’s recommendations. The manufacturer establishes maintenance plans about when and how long maintenance will take place, and provides maintenance guidance for each equipment and components. This type of routine maintenance scheduling does improve the reliability and reduce the failure risks of medical equipment to some extent. However, the regular inspection and replacement are inconsistent in most cases and are inefficient to cope with all types of failures, especially for those that are frequent and random [20].
As various monitoring technologies and tools have been developed during the last few decades, it was announced that the combination of preventive maintenance with monitoring data along with data analysis techniques would be the appropriate approach to predict equipment anomalies [21]. The Internet of Things (IoT), which integrates information of machine components through Internet by using modern information technologies, has emerged as a crucial technology to monitor the real-time status of targeted equipment and provide warnings in advance [22]. Particularly, the Internet of Medical Things (IoMT), where health data obtained from wearable devices and sensors to monitor real-time physical conditions [23], has received extensive attention. Currently, the development of IoMT is still at its early stage and most of the existing IoMT systems are focusing on improving the level of diagnosis that related to human body, rather than the medical equipment [24]. Likewise, it requires data-driven approaches to effectively store, preprocess and analyze the massive amount of log data that generated by IoT [25–27].
In this paper, we use machine-learning methods to predict CT anomalies based on the real-time condition data of CT equipment provided by the IoMT in West China Hospital of Sichuan University. In the CT system, parameters related to the condition of CT, such as oil temperature, anode voltage, daily arcing time and daily scan time, etc. are continuously monitored. By comparing different machine-learning models, we finally choose the one with the best prediction performance and improve the accuracy above 70% by adjusting parameters. Our research can significantly minimize the stagnation and losses, and improve the maintenance management. To the best knowledge of the authors, this is the first time that a sophisticated IoMT on large-scale medical equipment is developed and meanwhile, to be applied to investigate the anomaly of the medical equipment. Meanwhile, it is fairly new to combine the state-of-art machine learning models with the advanced monitoring tools in the medical field.
The rest of this paper is organized as follows: section 2 describes a typical CT system and the dataset we obtained from IoMT in the West China Hospital; section 3 demonstrates our research procedure; section 4 shows the results; conclusions are given in Section 5, with some perspectives and discussions on the future development.
2.1 The CT equipment
Figure 1 shows the diagram of a typical CT equipment diagram. In a scan process, the fan-shaped beam of X-rays emitted from the CT tube passes through the patient onto a number of digital detectors, which receive the X-rays and convert them into medical images. During this process, the X-ray tube generates a large amount of heat while only about 1% of electrical energy is effectively converted into X-rays [28]. As the overheating of CT tube may lead to the failure of the entire CT system, a cooling (cooling oil) system as shown in Fig. 2 is implemented to dissipate the heat [29]. Besides, the X-ray tube that works at high voltages requires a high-vacuum environment inside the tube. Arcing occurs when the required vacuum environment is broken and a conductive bridge between the cathode and the impurities is formed [30]. Although various anomalies of X-ray tubes have been discussed in the literature, the tube arcing is generally considered as the most typical and dangerous early sign of a CT equipment failure, as arcing is closely related to unstable performance of CT or breakdown of the tube [30]. Moreover, the voltage and current of the X-ray tube are also key factors that can reflect the arc of the tube [31]. Therefore, factors including inappropriate oil temperature, improper voltage, and unstable current, etc. should be evaluated when considering the anomalies of CT equipment.
2.2 Data Description
In this study, the continuously monitored real-time CT equipment status data from the IoMT of the West China Hospital is used to predict the anomaly of the CT equipment. This dataset contains the operational status data of 11 CT equipment. The features of the dataset include: Oil Temperature (\(OT\)), Anode Voltage (\(AV\)), Cumulative Tube Scanning Time (\(TST\)) and Cumulative Consumption of the Electrical Energy (\(CE\)). The appearance of arcing in the tube, which is closely related to the CT failure, is treated as labels. As the arcing data of only 3 CT equipment are available, we only consider the data of these 3 equipment (CT1, CT2, CT3) in this project. The details of the dataset are shown in Table 1. As shown in the Gantt chart in the table, missing data appears from time to time in the dataset due to IoMT system malfunction. Moreover, as there are a total of 33 observations in the arcing class, which is significantly smaller than that of 733 in the non-arcing class, the dataset is imbalanced. This will be considered in the model development.
Table 1 The details of the CT dataset from the IoMT
3.1 Data Preprocessing and Features Construction
As shown in Fig. 3, the raw observations from different sensors are obtained at a non-uniform frequency. We average the original observations of each sensor for that day and get the daily average data. In this way, the following features are obtained: the Daily Average Oil Temperature (\(AOT\)), Daily Average Anode Voltage (\(AAV\)), Daily Average Cumulative Tube Scanning time (\(ATST\)), and Daily average Cumulative Consumption of Electrical Energy (\(ACE\)).
The new features including Daily Tube Scanning Time (\(DTST\)) and Daily Consumption of Electrical Energy (\(DCE\)) are obtained by taking the first-order difference of \(TST\) and \(CE\) data. In addition, as the CT tube anomaly is closely related to the equipment operating current, the equipment current is obtained as a new feature based on the following equation:
$$I{\text{=}}\frac{{DCE}}{{AAV \cdot DTST}}$$
1
The CT equipment healthy state is also related to its idle time due to the cold emission phenomenon [9] which leads to ionization and arcing inside the idle X-ray tube. Based on this phenomenon, the new feature \(IDLE\) is created, which indicates whether the X-ray tube is idled in the past days. Besides, in order to improve the accuracy of the model, the derivation of the \(AOT\), \(AAV\), \(DTST\), \(DCE\), and data are obtained and denoted as \(AO{T_d}\), \(AA{V_d}\), \(DTS{T_d}\), \(DC{E_d}\), \({I_d}\) [32].
The Sliding Window algorithm is a method that has been widely implemented to predict future values, which constructs new features using the historical data of the previous days. In constructing the sliding window features, we follow these two rules:
(1) Extract the maximum or average values in the time window.
(2) Based on (1), if the value meets a certain condition, it is marked as 1. Otherwise, it is marked as 0.
Based on 3.1, the following features are obtained as shown in Table 2.
| Feature | Description | Window size | Time lag |
|---|---|---|---|
| \(AO{T_{Max}}\) | Maximum \(AOT\) data in the time window | \({n_1}\) | 1 |
| \(AA{V_{Max}}\) | Maximum \(AAV\) data in the time window | \({n_2}\) | 1 |
| \({I_{Max}}\) | Maximum data in the time window | \({n_3}\) | 1 |
| \(IDLE\) | \(IDLE=1\) if there is at least one day IoMT receives no data from the equipment in the time window. Otherwise, \(IDLE=0\). | \({n_4}\) | 1 |
| \(AO{T_{dMax}}\) | \(AO{T_{dMax}}=1\) If the maximum absolute value of \(AO{T_d}\) is greater than a threshold in time window. Otherwise, \(AO{T_{dMax}}=0\). | \({n_5}\) | 1 |
| \(AA{V_{dMax}}\) | \(AA{V_{dMax}}=1\) If the maximum absolute value of \(AA{V_d}\) is greater than a threshold in time window. Otherwise, \(AA{V_{dMax}}=0\). | \({n_6}\) | 1 |
| \({I_{dMax}}\) | \({I_{dMax}}=1\) If the maximum absolute value of \({I_{Max}}\) is greater than a threshold in time window. Otherwise, \({I_{dMax}}=0\). | \({n_7}\) | 1 |
| \(Arcin{g_w}\) | \(Arcin{g_w}=1\) if there is at least one day IoMT receives the observation of arcs in the tube from the equipment in the time window. Otherwise, \(Arcin{g_w}=0\). | \({n_8}\) | 1 |
3.2 Training and Testing Dataset Construction
As the observations with arcing labels are mainly from CT1, two frameworks are proposed to construct the training and testing datasets. As shown in Fig. 4, framework 1 performs stratified sampling on data from three equipment by proportion and then splits them into training and test set, while framework 2 only performs stratified sampling on the data of equipment CT1 by proportion and then add the data from the other two equipment with label 1 to the training set. Random oversampling is used to increase the size of the minority class [33]. In the process of splitting the training and the test set, the time window data of an arc will only be split in the training set or the test set.
3.3 Multivariate Time Series Classification Models
Models including Decision Tree (DT), Support Vector Machine (SVM), Logistic Regression (LR), Naive Bayesian (NB), K-Nearest Neighbor (KNN) are used in the prediction of anomalies in the multivariate time series dataset. The performance of those models are compared and the optimal one is selected.
We compare the evaluation measures using 5-fold cross-validation for each parameter combination. Through empirical analysis, the optimal window size parameters in Table 2 are obtained which result in the best model performance.
3.4 Performance Evaluation
Accuracy, Recall, Precision, and F1-score are used as the performance metrics of the classification models [34]. Accuracy is calculated by Eq. (2) to reflect the overall classification ability. Recall is calculated by Eq. (3), which is the fraction of true anomalies that are predicted as anomalies. Precision is calculated by Eq. (4), which is the true anomalies predicted as anomalies. F1-score is calculated by Eq. (5), which is the performance metric that considers both Recall and Precision simultaneously.
$${\text{Accuracy=}}\frac{{{\text{TP+TN}}}}{{{\text{TP+TN+FP+FN}}}}$$
2
$${\text{Recall=}}\frac{{{\text{TP}}}}{{{\text{TP+FN}}}}$$
3
$${\text{Precision = }}\frac{{{\text{TP}}}}{{{\text{TP+FP}}}}$$
4
$${\text{F1-score}}=2*\frac{{{\text{Recall}}*{\text{Precision}}}}{{{\text{Recall}}+{\text{Precision}}}}$$
5
where \({\text{TP}}\) is true positive, \({\text{TN}}\) is true negative, \({\text{FP}}\) is false positive and \({\text{FN}}\) is false negative.
Table 3 shows the performance metrics of the 5 models under the two frameworks with the optimal parameters, where the parameters shown in Table 4 are obtained from more than 59,000 parameter combinations. The models under Framework 2 are generally better than framework 1. Particularly, the performance metrics of KNN model under Framework 2 is the best, with the Accuracy, Recall, of 0.70 and 0.81 respectively. In addition, LR and NB models also perform well. The receiver operating characteristic (ROC) curves and the area under the curves (AUC) of the three models under Framework 2 are obtained as shown in Fig. 5. It is observed that the KNN model, with the highest (0.73) AUC value, has the best overall classification performance.
| Models | Accuracy | Recall | Precision | F1-score | |
|---|---|---|---|---|---|
| Framework 1 | DT | 0.82 | 0.45 | 0.28 | 0.26 |
| SVM | 0.94 | 0.08 | 0.46 | 0.12 | |
| LR | 0.62 | 0.80 | 0.23 | 0.31 | |
| NB | 0.64 | 0.86 | 0.31 | 0.37 | |
| KNN | 0.76 | 0.65 | 0.28 | 0.35 | |
| Framework 2 | DT | 0.73 | 0.61 | 0.33 | 0.40 |
| SVM | 0.86 | 0.11 | 0.56 | 0.19 | |
| LR | 0.71 | 0.71 | 0.31 | 0.41 | |
| NB | 0.75 | 0.67 | 0.33 | 0.42 | |
| KNN | 0.70 | 0.81 | 0.33 | 0.46 |
| \({n_1}\) | \({n_2}\) | \({n_3}\) | \({n_4}\) | \({n_5}\) | \({n_6}\) | \({n_7}\) | \({n_8}\) |
|
|
| |
|---|---|---|---|---|---|---|---|---|---|---|---|
| LR | 3 | 3 | 5 | 4 | 5 | 3 | 5 | 3 | 20 | 50 | 0.180 |
| NB | 3 | 5 | 5 | 4 | 4 | 5 | 5 | 3 | 30 | 40 | 0.180 |
| KNN | 5 | 5 | 5 | 4 | 5 | 3 | 4 | 3 | 30 | 40 | 0.216 |
The two most important features identified by the three models under Framework 2 are shown in Table 5. The results show that \(AA{V_{Max}}\) and \(AO{T_{Max}}\) are the most important features in models LR and KNN, while \({I_{Max}}\) and \(AA{V_{dMax}}\) are the most important features in model NB. This suggests that the raw features \(OT\) and \(AV\) are of primary importance in predicting the anomalies. Meanwhile, it also indicates that the reliability can be improved by properly operating the CT equipment, e.g., adjust the anode voltage slowly rather than quickly.
| LR | NB | KNN | |
|---|---|---|---|
| 1 | \(AA{V_{Max}}\) | \({I_{Max}}\) | \(AA{V_{Max}}\) |
| 2 | \(AO{T_{Max}}\) | \(AA{V_{dMax}}\) | \(AO{T_{Max}}\) |
Given the losses of hospitals and patients that, resulted from anomalies of medical equipment, predicting those anomalies in advance is of vital importance. In our work, a data-driven framework that preprocesses and analyzes time series status data obtained from the IoMT is proposed to predict anomalies of CT equipment. Specifically, seven new features are constructed through sliding time window based on the failure mechanism of CT equipment and then, two methods for splitting training and test sets are proposed, where random oversampling is applied to deal with imbalanced data. Our results show that the proposed framework is effective in predicting anomalies, with the accuracy of 0.70 and recall of 0.81 under the KNN model.
In this study, we demonstrate the efficiency of applying machine-learning methods to predict anomalies of medical equipment that is continuously monitored by the IoMT in the hospital. In practice, although large-scale medical equipment such as CT is critical for disease diagnoses and is extensively used every day, very few in-depth studies have been conducted to ensure its reliability during the operation. Regular maintenance strategies recommended by the manufactures show little efficiency in preventing unexpected failures. To the best knowledge of the authors, the work of this paper is a pioneering attempt to predict anomalies of large-scale medical equipment based on the medical equipment IoMT data. It enables the maintenance team to assess the reliability of equipment in real time and make proper maintenance decisions accordingly. In the future, the machine-learning methods combined with the IoMT technique that, flexible in incorporating unexpected failures of equipment, will show great potential in ensuring reliability of equipment and allocating equipment resources in the medical field.
There are some limitations that should be taken into account. First, monitored data containing only three CT equipment is available for model training, the result of which may not be extensively applied to other CT equipment. Second, with limited raw features, we construct the physical feature and \({I_d}\) by simply assuming that the CT equipment working in DC (direct current) circuits, where deviations may exist. For the future study, monitored data with high returned frequency that containing more state features are required to obtain a more accurate classification result. Moreover, it is highly recommended to establish IoMT that covering a wider range of medical equipment, by which more medical facilities could benefit from the data-based predictive-maintenance strategies.
In this study, we propose a novel framework that, integrates multivariate IoMT time series data, including preprocessing, feature construction, and classification. Based on that, we explore the application of machine-learning algorithms to predict anomalies in large-scale medical equipment.
The accuracy and the recall of the KNN reach 0.70 and 0.81, which has the best prediction performance in this case. With the IoMT built in the hospital continuously monitoring the state of large-scale equipment, it is efficient for medical facilities to utilize machine-learning methods to predict equipment anomalies in practice. In future works, with a higher quality of monitored data being available, we should consider more classification models and accordingly, improve the prediction accuracy.
CT: Computed Tomography; TJC: The Joint Commission; CM: Corrective Maintenance; PM: Preventive Maintenance; PdM: Predictive Maintenance; IoT: Internet of Things; IoMT: Internet of Medical Things; OT: Oil Temperature;AV: Anode Voltage; TST: Cumulative Tube Scanning Time; CE: Cumulative Consumption of the Electrical Energy; AOT: Daily Average Oil Temperature; AAV: Daily Average Anode Voltage; ATST: Daily Average Cumulative Tube Scanning time; ACE: Daily average Cumulative Consumption of Electrical Energy; DTST: Daily Tube Scanning Time; DCE: Daily Consumption of Electrical Energy; I: Current; DT: Decision Tree; SVM: Support Vector Machine; LR: Logistic Regression; NB: Naive Bayesian; KNN: K-Nearest Neighbor
Funding This study was supported by National Key Research and Development Program of China (2020YFB1711500), The 1·3·5 project for disciplines of excellence, West China Hospital, Sichuan University (ZYYC21004).
Role of funding sourceIt is used to pay the salaries of the paper participants.
Authors’ contributionChangxi Wang, Haopeng Zhou, Tong Wu, Haowen Liu and Kang Li analyzed the data, proposed the methods and wrote the main manuscript text. Qilin Liu, Yixuan Zhuo, Zhenlin Li and Jin Huang built the IoMT and acquired the data.
Conflicts of interest/Competing interests The authors declare that they have no conflict of interest.
Ethics approval This article does not contain any studies with human participants performed by any of the authors.- B. Wang (2012) Medical equipment maintenance: management and oversight. Synthesis Lectures on Biomedical Engineering 7: 1–85 https://doi.org/10.2200/S00450ED1V01Y201209BME045
- R. E. McClead and M. Brady (2016) Sentinel events/patient safety events. Pediatrics in Review 37: 448–450 https://doi.org/10.1542/pir.2015-0163.
- B. Wang et al. (2010) Evidence-based maintenance: part I: measuring maintenance effectiveness with failure codes. Journal of Clinical Engineering 35: 132–144 https://doi.org/10.1097/JCE.0b013e3181e6231e
- T. Yu, C. Zhu, Q. Chang, and J. Wang (2019) Imperfect corrective maintenance scheduling for energy efficient manufacturing systems through online task allocation method. Journal of Manufacturing Systems 53: 282–290 https://doi.org/10.1016/j.jmsy.2019.11.002
- S. K. Yang (2003) A condition-based failure-prediction and processing-scheme for preventive maintenance. IEEE Transactions on Reliability 52: 373–383 https://doi.org/10.1109/TR.2003.816402
- Y.-T. Tsai, K.-S. Wang, and H.-Y. Teng (2001) Optimizing preventive maintenance for mechanical components using genetic algorithms. Reliability Engineering & System Safety 74: 89–97 https://doi.org/10.1016/S0951-8320(01)00065-5
- D. Lin, M. J. Zuo, and R. C. M. Yam (2000) General sequential imperfect preventive maintenance models. International Journal of Reliability, Quality and Safety Engineering 7: 253–266 https://doi.org/10.1142/S0218539300000213
- M.-Y. You, H. Li, and G. Meng (2011) Control-limit preventive maintenance policies for components subject to imperfect preventive maintenance and variable operational conditions. Reliability Engineering and System Safety 96: 590–598 https://doi.org/10.1016/j.ress.2010.12.015
- J.-H. Jim and D. H. Park (2007) Optimal periodic preventive maintenance schedules with improvement factors depending on number of preventive maintenances. Asia-Pacific Journal of Operational Research 24: 111–124 https://doi.org/10.1142/S0217595907001139
- O. Ljungberg (1998) Measurement of overall equipment effectiveness as a basis for TPM activities. International Journal of Operations & Production Management 18: 495–507 https://doi.org/10.1108/01443579810206334
- A. Grall, L. Dieulle, C. Bérenguer, and M. Roussignol (2002) Continuous-time predictive-maintenance scheduling for a deteriorating system. IEEE Transactions on Reliability 51: 141–150 https://doi.org/10.1109/TR.2002.1011518
- M. C. Garcia, M. A. Sanz-Bobi, and J. d. Pico (2006) SIMAP: intelligent system for predictive maintenance: application to the health condition monitoring of a windturbine gearbox. Computers in Industry 57: 552–568 https://doi.org/10.1016/j.compind.2006.02.011
- K. Abdennadher, P. Venet, G. Rojat, J.-M. Rétif, and C. Rosset (2010) A real-time predictive-maintenance system of aluminum electrolytic capacitors used in uninterrupted power supplies. IEEE Transactions on Industry Applications 46: 1644–1652 https://doi.org/10.1109/TIA.2010.2049972
- S. M. Rezvanizaniani, J. Dempsey, and J. Lee (2020) An effective predictive maintenance approach based on historical maintenance data using a probabilistic risk assessment: PHM14 data challenge. International Journal of Prognostics and Health Management 5: https://doi.org/10.36001/ijphm.2014.v5i2.2237
- J. Lee (1997) Strategy and challenges on remote diagnostics and maintenance for manufacturing equipment. Annual Reliability and Maintainability Symposium, IEEE, Philadelphia, PA, USA. 368–370 https://doi.org/10.1109/RAMS.1997.571735
- H. W. Penrose (2009) Simple time-to-failure estimation techniques for reliability and maintenance of equipment. IEEE Electrical Insulation Magazine 25: 14–18 https://doi.org/10.1109/MEI.2009.5191412
- H. M. Hashemian (2011) Wireless sensors for predictive maintenance of rotating equipment in research reactors. Annals of Nuclear Energy 38: 665–680 https://doi.org/10.1016/j.anucene.2010.09.012
- C. R. Cassady, W. P. M. Jr, and E. A. Pohl (2001) Selective maintenance for support equipment involving multiple maintenance actions. European Journal of Operational Research 129: 252–258 https://doi.org/10.1016/S0377-2217(00)00222-8
- S. Panagiotidou and G. Tagaras (2008) Evaluation of maintenance policies for equipment subject to quality shifts and failures. International Journal of Production Research 46: 5761–5779 http://dx.doi.org/10.1080/00207540601182260
- A. B. Khalaf, Y. Hamam, Y. Alayli, and K. Djouani (2013) The effect of maintenance on the survival of medical equipment. Journal of Engineering, Design and Technology 11: 142–157 https://doi.org/10.1108/JEDT-06-2011-0033
- J. Maktoubian and K. Ansari (2019) An IoT architecture for preventive maintenance of medical devices in healthcare organizations. Health and Technology 9: 233–243 https://doi.org/10.1007/s12553-018-00286-0
- S. Li, L. D. Xu, and S. Zhao (2015) The internet of things: a survey. Information systems frontiers 17: 243–259 https://doi.org/10.1007/s10796-014-9492-7
- L. Syed, S. Jabeen, M. S., and A. Alsaeedi (2019) Smart healthcare framework for ambient assisted living using IoMT and big data analytics techniques. Future Generation Computer Systems 101: 136–151 https://doi.org/10.1016/j.future.2019.06.004
- S. Razdan and S. Sharma (2021) Internet of Medical Things (IoMT): overview, emerging technologies, and case studies. IETE Technical Review https://doi.org/10.1080/02564602.2021.1927863
- S. Gupta, R. Kambli, S. Wagh, and F. Kazi (2015) Support-vector-machine-based proactive cascade prediction in smart grid using probabilistic framework. IEEE Transactions on Industrial Electronics 62: 2478–2486 https://doi.org/10.1109/TIE.2014.2361493
- S. e. Z. Lashari, A. Takbiri–Borujeni, E. Fathi, T. Sun, R. Rahmani, and M. Khazaeli (2019) Drilling performance monitoring and optimization: a data-driven approach. Journal of Petroleum Exploration and Production Technology 9: 2747–2756 https://doi.org/10.1007/s13202-019-0657-2
- Z. Chi, T. Zhou, S. Huang, and Y.-F. Li (2020) A data-driven approach for the health prognosis of high-speed train wheels. Proceedings of the Institution of Mechanical Engineers, Part O: Journal of Risk and Reliability 234: 735–747 https://doi.org/10.1177/1748006X20929158
- A. A. Mohammed (2012) Improved cooling system for X-ray tube. Sudan University for Science & Technology
- L. Weston (2004) Cooling system for cooling an X-ray tube.
- L. Koch, A. Lesche, and W. Maring (1996) Fighting arcing and field emission in medical X-ray tubes. Xviith International Symposium on Discharges & Electrical Insulation in Vacuum, IEEE, Berkeley, CA, USA. 1077–1081 https://doi.org/10.1109/DEIV.1996.545531
- S. Tabakov (2018) X-ray tube arcing: manifestation and detection under quality control. MEDICAL PHYSICS international 6: 157–161 http://mpijournal.org/MPI-v06i01.aspx
- P. Schäfer and U. Leser (2017) Multivariate time series classification with WEASEL + MUSE. https://doi.org/10.48550/arXiv.1711.11343
- H. He and E. A. Garcia (2009) Learning from imbalanced data. IEEE Transactions on Knowledge Data Engineering 21: 1263–1284 https://doi.org/10.1109/TKDE.2008.239
- M. Irfan, Z. Jiangbin, S. Ali, M. Iqbal, Z. Masood, and U. Hamid (2021) DeepShip: an underwater acoustic benchmark dataset and a separable convolution based autoencoder for classification. Expert Systems with Applications 183: 115270 https://doi.org/10.1016/j.eswa.2021.115270
No competing interests reported.