Applying Neural Network-Based Approach to Sickle Cell Disease-Related Pain Classification

contributed to the Literature Review, edited the work and supervised the work. Israel Ogundele: Besides contributing to the Literature Review, wrote the initial draft of the work and also used Monte Carlo technique to generate the hypothetical random numbers (test data) for the work based on the expanded attributes. Olufunke Oladipupo: Reviewed the conceptual framework for the work. Provided cognate materials for the Literature Review, wrote the Conclusion section, and co-supervised the work. Each of the authors read and agreed with the contents.


Introduction
stated that computational Intelligence (CI) promised to advance the healthcare sector and clinical practice of disease management in diagnosis, treatment, prevention, prescription and optimization of the fast delivery to patient with these diseases. Akinwonmi (2011) confirmed Artificial Neural Network (ANN)'s connection/strength could be determined by the activation function which could be either linear or non-linear. Further, ANN consists of three layers (the input layer, output layer and hidden layer which is between the previous two layers. ANN's learning capabilities (supervised, unsupervised and reinforcement) are techniques used in learning. By adjusting the weight, the neural network adapts itself to learn and optimise to produce the desired output. Liu et al. (2006) affirmed ANN has been used in healthcare sector by applying the classification methods as ANNs identify the dataset features in order to accurately diagnose the nature of diseases, pains and sicknesses. According to Macintyre et al. (2010), Sickle Cell Disease (SCD), a systemic multiorgan disease that most commonly presents with painful vaso-occlusive crises, occurs either spontaneously or due to factors such as dehydration, infection, hypothermia and low oxygen tension. Jain and Gupta 2016); Xu et al., 2017) discovered that SCD, a hematological disorder, leads to blood vessel occlusion accompanied by painful episodes and even death. The three basic factors/blood components (RBC, MCH and Hb) that are affected in Symptomatic Sickle Cell Anaemia (SSCA) are: (1) Mean Corpuscular Haemoglobin (MCH), the average amount of haemoglobin found in red blood cells and measured in pictograms; (2) Red Blood Cells (RBCs), which carry oxygen. Normal RBC range in Males is 4.7 to 6.1 million cells per micro liter (cells/mcl) and in females is 4.2 to 5.4 million cells/mcl. People suffering from SSCA have RBC in the range 2.37-3.73 cells/mcl. RBCs of SCD patients have diverse shapes that reveal important biomechanical and bio-rheological characteristics, e.g., their density, fragility, adhesive properties, etc and (3) Hemoglobin (Hb or Hgb), the ironcontaining oxygen-transport metalloprotein in the RBCs, is measured in g/dl. A short summary of the variation in the values of these factors is given in Table 1. SCD becomes one of the most common severe lifethreatening haematological and monogenic disorders affecting millions of people worldwide. Affected blood is less able to carry oxygen and flow smoothly, which causes a host of health problems and a shorter lifespan. Piel et al. (2013;Kristiansen, 2014) reported that SCD occurs in people of African, Arabic and Indian racial backgrounds while countries in Equatorial Africa bear the greatest burden.  It was further stated that three-quarters of children who inherit SCD are born in sub-Saharan Africa, but most have no access to treatment and die before their fifth birthday. As shown in Table 2, worldwide, over 300,000 babies are born with SSCA yearly having sickle hemoglobin gene and this figure may increase to 400,000 by the year 2050. The usual disorder in an individual is SCD. Rees et al. (2010;Akinsete and Osu, 2017) have reported that 43 million people have sickle-cell traits and 4.4 million people have SCDs. Akinsete and Osu (2017) also estimated that 40 million Nigerians are carriers of this disease with over 150,000 infants born with SSCA. This situation translates to Nigeria having the highest record of SCD, where infant's death of this carrier is estimated to be 100,000 in Nigeria representing eight percent of her mortality rate. Case et al. (2018) submitted that SCD's management would social and cultural sensitivity of the practitioner's expert and experience due to the patient's challenging condition. Pain is a major issue in the care of patients with SCD. Dampier et al. (2014) found out that mechanisms behind pain and the best way to treat it have not been well understood. Vaso-Occlusive Crises (VOC) constitute SCD's painful episodes. Acute and chronic pains are mostly associated with adult patients, while acute pain is common in infants and children to classify the pain based on recent findings. On their own part, Macintyre et al. (2010) recommended that biopsychosocial assessment and multidisciplinary pain management might be required when treating patients with frequent, painful sickle cell crises. Telfer et al. (2014) concluded that SCD's acute painful crisis management remained unsatisfactory despite advances in the understanding and management of acute pain in other clinical settings. In corroborating, Lanzkron and Carlton (2015) discovered that the lack of a strong evidence base to guide the management/treatment of SCD-associated acute pain episodes have made it difficult for patients to receive high quality care outside of specialty centers. According to Poku et al. (2018) and Ginter et al. (2018), most hospitals and healthcare practitioners are using the traditional manual approach for management of patients with SCD, which can be time consuming and stressful to both patients and practitioners. Lanig et al. (2018) affirmed the daily increase in SCD would require more sufficient resources, such as healthcare professionals and practitioners, which are said to be inadequate. These inadequate healthcare workforce resources for SCD also often lack the required clinical experience while those working in the rural areas often lack the knowledge for clinical practice. Devi et al. (2013;Reader et al., 2017) reported ANN has been applied to SCsD for diagnosis, prediction and classification but has not been applied for the management of pain in SCDs.

Treatment of Sickle Cell Disease
Artificial Neural Network Optimization Jin et al. (2005) reported different researchers have compared various techniques such as Back-propagation, Simulated Annealing, ANN and Genetic Algorithm (GA) for optimizing processes in a network. While Simulated Annealing and GA have been proved to be global search techniques for optimization, Backpropagation algorithm is the mostly used for optimization techniques for training the neural network to find optimal solutions. While datasets are fed into the network through the input layer, the weight in the network is updated, by adjusting in an attempt to optimize the process and minimize the loss function, Sun et al. (2003) observed that best solutions are obtained in the area of the point which is more effective and consistent. In adjusting the weight, Agatonovic and Beresford (2000) had earlier stated that the interconnections of the nodes are strengthened while some are dropped (weakened) so that the neural network can output a better solution. Finally, the network training comes to halt when the best and optimal solution is obtained. Ibrić et al. (2012) opined that datasets' features could be classified into training and test datasets at a start of the training. Furthermore, the predictive uses test data while the training data is used to obtain optimal solution which changes the error, while the evaluation of the data is done by using both training and test data. The latter is done simultaneously. Table 3 shows the steps to follow in supervised training of network and usage. Macintyre et al. (2010) proposed that uncontrolled or unexpected pain would require a reassessment of the diagnosis and consideration of alternative causes for the pain (e.g., new surgical/medical diagnosis, neuropathic pain). Equally important is the fact that multiple outcome measures are required to adequately capture the complexity of the pain experience and how it may be modified by pain management interventions. Pain management in SCD is challenging due to both the frequent crisis being faced by the patient and the inability of the healthcare practitioner to quickly identify the pain as the RBC disorder constitutes the cause of the painful complication in SCD as ascertained by Case et al. (2017). Ballas (2005) had earlier classified pains in SCD as chronic pain, acute pain, neuropathic pain or mixed pain. These are unpredictable and can occur at any time. Chronic pain is the outcome of the frequent acute pain that has not been properly taken care of. The pain can be experienced between three months and more.  (Ibrić et al., 2012) Training of the network Data is presented to the network Network computes an output Network output is compared to desired output Network weights are modified to reduce error Usage of the network Present new, unseen data to the network Network computes an output based on its training Some of the symptoms are achy, frequent in nature and this happens in a pathophysiologic events. Acute pain can occur throughout the life of the SSCA patient. The pain can be so sudden leading to Vaso-Occlusion (VOC) crisis and can cause damage to the organ which can sometimes lead to death if not properly managed. Neuropathic pain occurs as a result of wound or dysfunction in the body and is associated with SCD which can be triggered by harmful or deadly things. Dampier et al. (2017) corroborated Ballas (2005)'s classification and expressed that SCD-related pain is associated with increased morbidity, mortality and high health care costs. Dampier et al (2017), in giving a common set of criteria for classifying chronic pain associated with SCD, concluded such classification would enhance SCD pain research efforts in epidemiology, pain mechanisms and clinical trials of pain management interventions and ultimately improve clinical assessment and management by adopting a Proforma for Pain Assessmentas given in Fig. 1 (Howard and Telfer, 2015).

Pain in Sickle Cell Disease
Due to the challenges in identifying and treating the painful episodes of the SCD which occur from offspring and can continue throughout the lifespan of that patient, frequent pain of acute nature requires quality healthcare and attention by the practitioners. Macintyre et al. (2010;Case et al., 2017;Howard and Telfer, 2015) recommended that the Acute Painful Crisis (APC), an episode of pain, usually of abrupt onset, in severe cases would require hospital treatment with opioid analgesia. While parenteral corticosteroids appear to reduce the duration of analgesia requirements and length of hospital stay, without major side effects, during sickle cell crises, one of the most important complications of APC is Acute Chest Crisis (ACS). In their submissions, Macintyre et al. (2010), Howard and Telfer (2015) and Case et al. (2017) identified monitoring for development of ACS with regular assessment of respiratory rate, oxygen saturation and daily examination of the chest as an essential part of the management of APC. There is insufficient evidence to suggest that fluid replacement therapy reduces SCD-associated pain. Hydroxyurea is effective in decreasing the frequency of acute crises, life-threatening complications and transfusion requirements in SCD.

The Sickling Phases
Hemoglobin has a rope-like structure -the sicklewhich is a trait in SSCA. Molecules of the hemoglobin are put together to form fibres and then aggregated into twisted pairs. US_NIH (2014) affirmed SSCAhemoglobin consists of four sickling phases: HbS, P + chain, deoxy+bs and polymer while each of the hemoglobin molecules is called heme group. Haemoglobin polymerisation leads to abnormal sickleshaped erythrocytes' rigidity to disrupt blood flow in small vessels. Figure 2 shows the sickling phases (Howard and Telfer, 2015).
The most prevalent SCD genotypes associated with the most severe clinical manifestations include homozygous hemoglobin SS (HbSS) and the compound heterozygous conditions hemoglobin Sβ0-thalassemia (HbSβ0-thalassemia), hemoglobin Sβ+-thalassemia (HbS β +-thalassemia) and hemoglobin SC disease (HbSC). Vaso-occlusion (central to the pathophysiology SCD) leads to distal tissue ischaemia and inflammation, with symptoms defining the acute painful sickle-cell crisis. The importance of chronic anaemia, haemolysis and vasculopathy has been established. Nevertheless, repeated sickling and ongoing haemolyticanaemia, even when subclinical, lead to parenchymal injury and chronic organ damage, causing substantial morbidity and early mortality. Consequences of sickling include destruction to the membrane and cytoskeleton, removal in the RBC, red cell dehydration and impaired anti-oxidant mechanisms. Figure 3 shows the sickle cell crisis which causes obstruction of blood flow and pain. Patients with ANY of the below should be referred for medical review Chest Pain, Shortness of breath, Hypoxia (oxygen saturation <94% Fever/rigors (Temp>38°C). Hypertension (BP <90/60) Tachycardia>1100 (even after pain has settled following analgesia) Raised respiratory rate of >20(even after pain has settled following analgesia) New neurological symptoms, headache, confusion, numbness of limbs Abdominal pains Priapism (persistent erection) Pregnancy Visual loss or bleeding in the eye PAR>four Concerns from the nursing learn about the patients' clinical condition.  Lanzkron et al. (2010) reported SCD causes intermittent and recurrent acute pain episodes, referred to as 'Vaso-Occlusive Crises' (VOC), as a result of vasoocclusion. Furthermore, most adults and infants experienced these painful episodes (being the most common with people having SSCA). VOC occurs when there is coagulation of the RBC and this can lead to severe injuries or damage to the organ of the body which is the most common in the complication. The SSCA's severe complication is responsible for Emergency Department (ED) and healthcare sector for SCD patient to receive quick treatment with high quality and outmost care to save life. Hereditary genetic disorder, also found in SCD, results in the predominant production of an abnormal/mutant hemoglobin called Hemoglobin S (Hb S) in RBCs. Persons carrying one normal hemoglobin gene and the S gene are known as having sickle trait (AS) and are usually healthy and haematologically normal. Sickle shaped red cells, formed when hemoglobin offloads oxygen to the tissues, are more rapidly removed from the circulation by the body (a process called hemolysis) leaving much less hemoglobin circulating (anaemia). Complication occurrences can lead to VOC. Every SCD patient experiences this VOC during his/her life span and if the person is not being taken care of during the time of the crisis, it might lead to death.

Materials and Methods
ANN's ability to learn can be used to implement the algorithm capable of learning and optimizing processes involved in the management of pain in SCD patient. Data are collected by hospitals and healthcare centers having records of the entire SCD patients that have been diagnosed and treated. With this data we can determine a patient that has greater risk by processing and analysing the data collected. Neural Network can help to process and analyze patients with sickle cell and those that need immediate attention. The use of ANN can help predict the best practice in management of pain during crisis based on the symptoms. Elsalamony (2016) presented an algorithm to identify healthy and unhealthy sickle cell patient using ANN as structured in Fig. 5. This was a neural network which has input layer, output layer and hidden layer (between input and output layers), but not trained to optimize the processes in the management of pain in SCD.

Frequency of observation (to assess oxygen saturations on AIR please remove oxygen for 5 minutes)
Pain score 0, Ask the patient to self-assess, 0 represents an absence of depression/anxiety/stress and 10 represents a very high level of depression/anxiety/stress. If record score is consistency>7 over 48 hour, refer to health psychology learn.

More score
Sedation score-Sedation can precede respiratory depression  Methods/ANN data for prevention and guidelines for manual approaches to sickle cell therapy. diseases in Sickle Cell. Rahmat et al.
Self-Organizing Map The paper classified normal and abnormal System developed could not optimize the (2018) Neural Network. RBC of the sickle cell diseases in digital process of pain management in SCD image using self-organizing map neural network.
From Table 4, various researchers acknowledged the fact that there is a demonstrated need for management of pain in SCD. Equally and importantly accepted is the need to optimize the processes of the pain management for better healthcare services.

Training Data for Sickle Cell Disease
Currently, there is no standardization for pain management in SCD. This paper attempts to develop a neural network model capable of promoting higherquality care by optimizing the processes in managing sickle cell patients during pain-induced crisis. This will help to improve the quality of life of the patient with attendant reduction of unnecessary spending, patient illness and pressure for the healthcare practitioners in terms of emergency cases they need to attend to per time. The SCD attributes adapted from Khalaf et al. (2016) in Table 4 were enhanced for consideration with Age, Educational Background and Location items. Monte Carlo Random Number Generation technique has been used in this research to generate dataset using the fifteen attributes of Table 5. This approach developed a scientific framework that facilitates hypothetical generation of a big dataset without necessarily going through the time-consuming Ethical Approval Committee process of Institutional Review Board. Educational Background Highest educational background of the patient to know the level of literacy 4. Haemoglobin The protein found in the Red Blood Cell (RBC) that carries oxygen to every part of the body. 5.
Location Geographical area and address where the patient lives. 6.
Mean corpuscular volume (MCV) The measure of the size of the red blood cells in the body of the patient. 7.
Platelets (PLTS) Thrombocytes: refer to components of blood whose function is to stop bleeding. 8.
Neutrophils count (RETIC A) Real number of White Blood Cells (WBC) present in the patient. 10.
Reticulocyte count (RETIC %) Measures the rate at which reticulocytes are made in the bone marrow and enter the bloodstream. 11.
Alanine Aminotransferase (ALT) test The blood test that checks for liver damage. It's an enzyme mostly in liver and kidney cells 12.
Body Bio Blood (BIO) BodyBio wellness Report is a revolutionary report that lets you get the most from the companion blood test. 13.
Fetal Hemoglobin (HbF) The RBC that carries oxygen round the body. 14.
Bilirubin Helps to find the cause of health conditions, like jaundice, liver disease and anaemia. 15.
Lactate Dehydrogenase (LDH) An enzyme involved in the energy production that is found in almost the body's entire cell.   Figure 6 presents the neural network to manage the pain in sickle cell patient to be able to determine between normal and abnormal patient in the processes, using the attributes in Table 5.
About 515 datasets were generated using Monte Carlo a Simulation Techniques to Generate Random Number for items in Table 5.

Results
For the purpose of the work we classified pain into four: low acute pain, severe acute pain, low chronic pain and severe chronic pain (crisis). Our target output percentages are as presented in Table 6.
Our results using Figs. 6 and 7 are presented in Table 7. To validate the system, we have a training dataset and testing dataset on classifier model. The evaluation of the neural network was based on some certain parameters such as sensitivity, specificity, precision, the F1 score, Youden's J statistic and the classification accuracy. Table  8 shows the formula to evaluate the performance.
Where TP, TN, FP and FN stand for true positive, true negative, false positive and false negative respectively for the evaluation performance. Backpropagation method was used for training the datasets. The activation function was determined by the weight of the network, the gradient of the loss function fed into the network to the back-propagation to update the weights of the function in order to reduce the loss function. This was done using a supervised form of learning in ANN. The ANN could learn from the datasets and be able to recommend the best processes in management of pain of sickle disease patient.

Discussion
Back-Propagation Neural Network (Multilayer Perceptron) with supervised learning model has been applied in SCD pain management processes in promoting higher-quality care. The datasets features of the SCD patients were used to train the neural network according to the pain encountered in identifying and treating the patient as fast as possible. Performance evaluation metric of the training datasets would help determine the accuracy and effectiveness of the Neural Network for utmost result.

Conclusion
Sickle cell disease includes a group of inherited disorders of haemoglobin production. Haemoglobin S polymerises when deoxygenated, causing rigidity of the erythrocytes, blood hyperviscosity and occlusion of the microcirculation with resultant tissue ischaemia and infarction. Literature has ascertained that haemoglobin polymerization which leads to erythrocyte rigidity and vaso-occlusion is central to the pathophysiology of SCD. Equally established is the importance of chronic anaemia, haemolysis and vasculopathy. Clinical management is basic and few treatments have a robust evidence base. This paper provides a solution to SCD pain management during VOC. Painful episodes can easily be managed by the healthcare sectors to serve as a great relief. The development of this framework would help solve some of the challenges that are faced in the healthcare system in SCD-related pain management.

Funding Information
No funding was involved in this research work as it forms an extract from the second author's postgraduate degree programme pursuit being supervised by the first author.

Author's Contributions
Zacchaeus Omogbadegun: Originated and prepared the conceptual framework for the work. He wrote the Abstract, contributed to the Literature Review, edited the work and supervised the work.
Israel Ogundele: Besides contributing to the Literature Review, wrote the initial draft of the work and also used Monte Carlo technique to generate the hypothetical random numbers (test data) for the work based on the expanded attributes.
Olufunke Oladipupo: Reviewed the conceptual framework for the work. Provided cognate materials for the Literature Review, wrote the Conclusion section, and co-supervised the work.
Each of the authors read and agreed with the contents.

Ethics
To prevent ethical issues that may arise after the publication of the manuscript, Monte Carlo Random Number Generation technique has been used in this research to generate dataset using enhanced fifteen attributes. This approach developed a scientific framework that facilitated hypothetical generation of a big dataset without targeting a particular patient or necessarily going through the usual time-consuming Ethical Approval Committee process of Institutional Review Board. We have also run the manuscript through a plagiarism/similarity check (Turnitin) to ensure the originality and uniqueness of the research work. We have ensured every item in the References List has been properly cited and vice versa. In addition, we have made a direct and substantial contribution to the research.