Article Preview
TopIntroduction
A fetus in the mother’s womb is protected from the pathological threats since it is not directly accessible. Physicians thus need some ancillary means to observe the fetal heart rate (FHR) which is a source of information about the fetal health. Since 1960s cardiotocography (CTG) is the most preferable means of assessing the well-being of a fetus. The readings obtained from CTG may give indication about the onset of a pathological condition such as hypoxia or heart disease, which if identified early may help save the fetus from life-long disability or even death.
Visual interpretation of CTG is not unambiguous. Though the sensitivity for detection of pathological cases is acceptable, the specificity for suspicious cases is still not up to the mark. Confirmation in such cases require invasive methods such as pH balance test using fetal scalp blood which is not possible all the time. If a fetus is identified as hypoxic physicians usually recommend a C-section which may not be necessary. This increases sensitivity at the cost of specificity.
Peri-natal mortality is highly prevalent in developing countries in India. Occurrence of reported stillbirth in rural India are the highest among the countries of South East Asia such as Thailand, Myanmar and Sri Lanka (Shah, Pratinidhi, & Bhatlawande, 1984). In India, according to National Fetal Health Survey-3 (NFHS-3) (2005-06) peri-natal mortality and stillbirth rates are 4.85% and 1.92% respectively. But this may be a gross underestimate as many still births go unreported. Nearly 60% of these deaths are stillbirths and many of them are preventable. Leading causes of fetal demise are found to be asphyxia, unexplained stillbirths, intrauterine growth retardation, congenital anomaly etc. (Nishat, 2018).
Most of these stillbirths could be prevented with adequate antenatal and peri-natal care. Primary health care providers are responsible for late referral because they contribute to suboptimal care by failing to recognize high-risk cases on time. These include failure to manage high risk cases, delay/ error in labor management etc. (Nishat, 2018) by the clinicians. Availability of equipment such as CTG machines in health care centers would help in early detection of fetal condition and hence timely intervention.
Health care facilities in remote areas cannot provide or afford fetal monitoring devices as they are too expensive. Even if CTG machines are available, most of them are not automated, leading to visual interpretation by the physicians. Also, in many such areas a physician is not always be available during an emergency; in such case it may not be possible for other health care professionals to interpret the CTG. In such situations an automated yet cost-effective fetal monitoring device might be able to save the life of a fetus.
Machine learning approach is necessary, as it learns directly from the data. Most of the learning based tasks in medical diagnosis such as interpretation of the status of the fetus from the CTG, need to be supervised, i.e. the data needs to be manually annotated (Cömert, Şengür, Budak, & Kocamaz, 2019). Creating a gold-standard in this case is a tedious job, requiring the opinions of a number of experts, effective annotation mechanism and time.
Taking the opinions of multiple experts is crucial in reducing the effect of inter and intra observer variations. In the evaluation of CTG, high interpretation skill is needed as annotation results may vary depending upon the experience of the experts. In fact, it was noted that the same observer was providing different annotation to the same data at a different time. Thus, to create a gold-standard multiple annotations need to be collected and merged. The best way to achieve this goal is by proposing machine learning approaches to support the interpretation of various features of CTG and ultimate detection of the state of the fetus.
The contribution of this paper involves comparison of five different machine learning algorithms viz. FURIA, Random Forest (RF) and Multilayer perceptron (MLP), Naïve Bayes and SVM for the classification of the Baseline value of fetal heart rate. The statistical analysis of the end result reveals the machine learning algorithm best suited for our purpose.
Fuzzy Unordered Rule Induction Algorithm (FURIA) is considered one of the best fuzzy classification rule learning algorithm when it comes to medical decision making due to its high rate of accuracy (Das, Roy, & Saha, 2015a).