In this section, a few recent works in the field of big data analytics in healthcare are briefly presented.
Ojha et al.[19]presented an insight into how big data analytics tools like Hadoop can be used with healthcare data. They discuss how meaningful information can be extracted from EHR (elec- tronic health record). For this work, they conducted their experiments on the healthcare data obtained from central India’s major government hospital, Maharaja Yeshwantrao Hospital (M.
Y.) situated in Indore, Madhya Pradesh, India. This hospital generates a large amount of data every day, which the authors suggest to store in EHR. A unique number is assigned to every patient whose data is stored here. With this number, information about any patient can be accessed more easily and quickly. Also, the data warehouse, EDH (enterprise data hub), can be used to store this data. Different data mining techniques such as classification, clustering, and association can easily be performed on this data directly. This implies a steady and efficient medical big data analysis technique.
Koppad et al.[20]introduced an application of big data analytics in the healthcare system with an aim to predict COPD (chronic obstructive pulmonary disease). They have used the Decision Tree technique, a data mining technique to perform COPD diagnosis in an individual patient. The Aadhar number is used to refer to the patient’s details that are stored in a centralized clinical data repository. As the Aadhar number is unique, it links to the treatments given to the patient in dif- ferent hospitals and also about the doctors in charge. The authors claim an encouraging accuracy in diagnosing COPD patients and the efficacy of the proposed system through experimental results.
48 CHAPTER 3 BIG DATA ANALYTICS IN HEALTHCARE: A CRITICAL ANALYSIS
Patel et al.[21]discuss the reasons behind the consideration of data in healthcare and the results of various surveys to demonstrate the influence of big data in healthcare. They also present some case studies on big data analytics in healthcare industries. Different tools for handling big data problems are discussed.
Simpao et al.[22]focuses on big data analytics in the anesthesia and healthcare units. They also focus on visual analytics, which is the science of analytical reasoning simplified by interactive visual interfaces. This can also assist with the performance of cognitive activities concerning big data. The Anesthesia Quality Institute, and the Multicenter Preoperative Outcomes Group have led significant efforts to gather anesthesia big data for outcomes research and this also aids quality improvement. They suggested that the efficient use of data combined with quantitative and qualitative analysis to make decisions can be employed to big data for excellence and performance enhancement. For instance, sev- eral important applications are clinical decision support, predictive risk assessment, and resource management.
Jokonya et al.[23]propose a big data integration framework to support deterrence and control of HIV/AIDS, TB, and silicosis (HATS) in the mining industry. The link among HIV/AIDS, TB, and sil- icosis is the focus in this work. The authors claim that their proposed approach is the first one to use big data in understanding the linkage between HATS in the mining industry. The proposed big data frame- work addresses the needs of predictive epidemiology, which is important in forecasting and disease control in the mining industry. They suggest the use of a viable systems model and big data to tackle the challenges of HATS in the mining industry.
Weider et al.[24]introduce a Big Data Based Recommendation Engine for early identification of diseases in the modern health care environment. A classification algorithm, Naı¨ve Bayes (NB), is used to build this system. This algorithm runs on top of Apache Mahout and it advises the health conditions of users, readmission rates, treatment optimization, and adverse occurrences. The proposed work fo- cuses on analyzing and using new big data methodologies. The proposed approach is a very efficient one in the sense that once the disease is identified, it will be easy to deliver the correct care to the pa- tients and in this way, the average life expectancy of people can be increased if they are given suitable care from the early stages.
Chrimes et al.[25]propose a framework built to form a big data analytics (BDA) platform via the use of real volumes of healthcare big data. The existing high-performance computing (HPC) architec- ture is utilized with HBase (NoSQL database) and Hadoop (HDFS). The generated NoSQL database was imitated from metadata and inpatient profiles of the Vancouver Island Health Authority’s hospital system. A special modification of Hadoop’s ecosystem and HBase with the addition of “salt buckets” to ingest was utilized. The authors claim that data migration performance requirements of the proposed BDA platform can capture large volumes of data while decreasing data retrieval times and its associ- ations to innovative processes and configurations.
Chawla et al.[26]propose a personalized patient-centered framework, CARE. The proposed sys- tem serves as a data-driven computational support for physicians evaluating the disease risks facing their patients. It has the ability of early caution indicators of possible disease risks of an individual, which can then be converted into a dialogue between the physician and patient and this will aid in patient empowerment. CARE can be utilized in full potential to explore broader disease histories, rec- ommend previously unconsidered concerns, and facilitate discussion about early testing and preven- tion, as well as wellness strategies that may be more recognizable to the individual and easy to implement.
49 3.5 RECENT WORKS IN BIG DATA ANALYTICS IN HEALTHCARE DATA
McGregor in[27]discusses the benefits and effectiveness of the use of big data in neonatal intensive care units. He claims that it will lead to earlier discovery and deterrence of a wide range of fatal medical conditions. The capability to process multiple high-speed physiological data streams from numerous patients in numerous places and in real time could considerably improve both healthcare competence and patient outcomes.
Fahim et al.[28]propose ATHENA (Activity-Awareness for Human-Engaged Wellness Applica- tions) to plan and assimilate the association between the basic health needs and suggest the human lifestyle and real-time recommendations for wellbeing services. With this system, their motive is to develop a system to encourage an active lifestyle for individuals and to suggest valuable interferences by making comparisons to their past habits. The proposed system processes sensory data through an ML (machine learning) algorithm inside smart devices and exploits cloud infrastructure to decrease the cost involved. Here, big data infrastructure is employed for huge sensory data storage and fast retrieval for recommendations.
Das et al.[29]proposed a data-mining-based approach for the classification of diabetes mellitus disease (DMD). They applied J48 and Naı¨ve Bayesian techniques for the early detection of diabetes.
Their proposed model is elaborated in consecutive steps to help the medical practitioner to easily explore and recognize the discovered rules better. The dataset used is collected from a college med- ical hospital as well as from the online repository. Further practical applications are based on the proposed approach. The PSO (particle swarm optimization) based approached can also be employed for this classification task. One such type of classification technique can be found in[30]. This tech- nique is a PSO-based evolutionary multilayer perceptron, which is trained using the back propagation algorithm. Some other advanced techniques such as the one proposed in[31]can also be adopted for the classification task. This technique[31] is based on the De-Bruijn graph with the MapReduce framework, and it is used for metagenomic gene classifications. The graph-based MapReducing ap- proach has two phases: mapping and reducing. In the mapping phase, a recursive naive algorithm is employed to generate K-mers. The De-Bruijn graph is a compact representation of K-mers that finds out an optimal path (solution) for genome assembly. The authors utilized similarity metrics for find- ing similarity among the DNA (De-Oxy Ribonucleic Acid) sequences. In the reducing phase, Jaccard similarity and purity of clustering are applied as dataset classifiers to classify the sequences based on their similarity. The experimental results claim this technique is an efficient one for metagenomic data clustering.
But it is also important to discuss the possible security threats that may arise while transferring med- ical images and data over the internet. To deal with privacy and copyright protection of such a huge amount of medical data, we need more robust and efficient techniques. These different techniques should be studied and tested empirically for finding the most efficient technique that is easy to imple- ment and can provide optimal protection. Such different techniques are thoroughly discussed in [32, 33]. Different challenges that might be faced during the analysis of medical big data also need to be addressed. A very good study on this can be found in[34].
In summary, above are some noteworthy contributions towards big data analytics in healthcare data.
These works contain some innovative ideas for utilizing big data analytics in healthcare data to extract new valuable information and thereby discover innovative ways to deal with different serious diseases.
To deal with healthcare big data, there should be a sound architectural framework and then arises the need for some big data analytics tools. These tools are briefly introduced, along with their advantages and disadvantages in the below section.
50 CHAPTER 3 BIG DATA ANALYTICS IN HEALTHCARE: A CRITICAL ANALYSIS