Home > Free Essays > Tech & Engineering > AI > Artificial Intelligence in Healthcare Delivery
Rate

Artificial Intelligence in Healthcare Delivery Report

Exclusively available on IvyPanda Available only on IvyPanda
Updated: Dec 3rd, 2020

Abstract

The human eye is viewed as the most powerless organ to be influenced by diabetes. Diabetic retinopathy is a complex phase of diabetes and is separated into two stages. Invasion in the optical layer is the fundamental explanation behind the disintegration of visual impairment. Over the years, therapies for diabetic retinopathy treatment concentrated on controlling side effects and halting the weakening of visual impairment.

However, with the advancement of technologies in eye medical procedure, treatment plans try to restore the patient’s vision to full recovery. This report presents the status of AI in healthcare delivery and the motivations of deploying the technology in human services, information types analysed by AI frameworks, components that empower clinical outcomes and disease types. The benefits of AI have been discussed in different works of literature.

AI can utilise advanced algorithms to evaluate components from a vast volume of human service information, and use the acquired knowledge to help clinical practice. The technology can be upgraded with learning and self-redressing capacities to improve its precision based on patient diagnosis. AI frameworks assist doctors by giving current restorative data from diaries, course readings and clinical practices to advise appropriate patient care. Consequently, the AI framework can decrease unavoidable demonstrative and remedial blunders in clinical exercise.

Objectives of the AI System

Artificial intelligence (AI) is expanding with many advantages for economies, social orders, networks, and people. AI innovations improve efficiency and making new items and administrations. These innovations are connected in areas of retail, assembling, diversion, pharmaceuticals, training and transport. In simple terms, artificial intelligence means to copy human psychological capacities.

It is conveying a change in perspective to social insurance, controlled by expanding accessibility of human services information and quick advancement of examination procedures. AI can be connected to kinds of human service information. Prevalent AI procedures incorporate machine-learning strategies for classified information, for example, the traditional vector machine, neural system and deep learning for unstructured data (Mirsharif, Tajeripour, & Pourreza, 2013).

Significant ailment territories that use AI devices incorporate cancer, nervous system science and cardiology. As a result, artificial intelligence supports early recognition, diagnosis, forecast and visualisation evaluation. In an endeavour to beat impediments inborn in automated diagnosis, specialists have done projects that reenact human thinking. Expectations that such a technique would prompt significant benefit have not been reported, yet many challenges have been explained.

Methodologies have been created to confine the number of speculations that a program must consider and fuse pathophysiologic thinking. The technology allows an application to examine cases that affect the introduction of another anomaly. Models encapsulating such thinking can clarify their decisions in medical terms. Despite these advances, further research and formative endeavours should be conducted to perfect AI technologies.

Opportunity in AI Proposed Project

The capacity to represent the human personality and conduct complex evaluation is named artificial intelligence. Given the difficulties and unpredictability related to AI, numerous analysts directed their concentration toward narrow AI, which is the capacity to conduct specific assignments. AI procedures have sent huge waves over healthcare insurance, fuelling a functioning view of whether AI specialists will displace human doctors. Although machines cannot replace human doctors, AI can help doctors settle on better clinical choices or even supplant human judgment in certain useful territories such as radiology (Yazid, Arof, & Isa, 2012).

The expanding accessibility of human service information and quick improvement of big data strategies supports the ongoing applications of artificial intelligence in healthcare services. Guided by important clinical inquiries, AI systems can explain clinical challenges and diagnoses. The breakthrough in science would improve clinical decision-making. This report presents the status of AI in healthcare delivery and the motivations of deploying the technology in human services, information types analysed by AI frameworks, components that empower clinical outcomes and disease types. The benefits of AI have been discussed in different works of literature. AI can utilise advanced algorithms to evaluate components from a vast volume of human service information, and use the acquired knowledge to help clinical practice (Raja & Gangatharan, 2015).

The technology can be upgraded with learning and self-redressing capacities to improve its precision based on patient diagnosis. AI frameworks assist doctors by giving current restorative data from diaries, course readings and clinical practices to advise appropriate patient care. Consequently, the AI framework can decrease unavoidable demonstrative and remedial blunders in clinical exercise.

To justify the AI Powered Solution

Quality HealthCare Delivery

Computer-based intelligence can be used in arranging and asset assignment in health institutions and social administrations. For instance, the IBM Watson solution improves cost-effectiveness. The AI engineered system matches people with an appropriate physician that addresses their issues based on the assigned financial plan. It additionally structures personal health plans and offers insights for resource management. Artificial intelligence has been deployed to improve the client experience. The application recognises patient tensions before a visit, give data on interest, and furnish clinicians with information to deliver proper treatment.

Clinical Study

Artificial intelligence can be utilised to examine and recognise designs in substantial and complex datasets (Weng, Reps, Kai, Garibaldi, & Qureshi, 2017). AI technology has improved task delivery, enhance quick asset deployment for emergencies. Based on its advantages, AI has been used to examine the logical literature for appropriate investigations and to consolidate various types of information for a drug study. Artificial intelligence frameworks utilised in human services could be significant for restorative research by coordinating appropriate patients for clinical studies.

Healthcare Delivery

Computer-based intelligence supports human diagnosis. Utilising AI to dissect clinical information, examine distributions, and proficient rules could educate the physician on the choice of treatment. Conceivable employments of AI in clinical consideration include clinical imaging, echocardiography, testing for neurological conditions and surgery.

Healthcare Applications

A few applications that utilise AI to offer customised clinical evaluations and home consideration guidance has been invented. The application, Ada Health Companion uses artificial intelligence to provide accurate diagnoses. Data instruments run by artificial intelligence aid the administration of unending ailments. AI technologies assist physicians through communications with patients to give customised data and guidance concerning treatment therapy, drug use, and exercise. Government-subsidized and business activities are investigating courses in which AI could be utilised to control mechanical frameworks and applications to help individuals living with memory challenges. As a result, AI in human services and research, conceivably decrease requests on caregivers and family support groups.

General Health

AI can be utilised to help early recognition of disease epidemics. Artificial intelligence has been employed to foresee antagonistic medication responses, which are assessed to cause up to 6.5 per cent of medical clinic affirmations in the UAE. Based on these justifications for proposal, this report will discuss the importance of artificial intelligence for screening diabetic retinopathy.

Links with Strategic Goals of the Sponsoring Organization

It is notable that eye defects and complications can be analysed through non-obtrusive imaging of the retina. Early screening for diabetic retinopathy is critical because it prevents visual impairment. Such detection additionally provides information about other cardiovascular sickness caused by primary diseases. The requirement for such screening and the need for accurate examination motivate the objective of this report.

Routine imaging for detection utilises the exceptionally planned optics of a ‘fundus camera,’ with pictures taken at various angles. Appraisal of the photograph requires specialists. However, technology development has introduced advanced digital recording of retina photographs (Niemeijer, Abràmoff, & van Ginneken, 2009). The transformational advancement in digital retinal picture examination, utilising various branches of AI has been demonstrated. For example, deep learning model improves imaging outcomes.

Objectives of the Report

Financial, social, and therapeutic expenses of diabetes are substantial medical issues. It is troublesome that cardiovascular difficulties concerning diabetes affect the patient’s recovery phase. Therefore, diagnoses, classification of diabetes and treatment therapy are essential to research issues in clinical studies. An imperative complexity of diabetes is called diabetic retinopathy. Diabetic retinopathy is a complication that causes visual interference.

However, first examination and treatment are critical to prevent visual impairment conditions. Thus, mass screening of diabetic patients is profoundly alluring. Computerised image assessment can be a reliable tool in aiding diagnoses comprising diabetic retinopathy and specific forms of muscular deterioration (Lu et al., 2018). The first approach examines and quantifies symptoms of the ailment like human diagnosis. The second approach is to enable the system to establish its evaluation using different samples.

Thus, incorporating AI in the identification of retina disease diagnosis will replace these traditional techniques and provide accurate results. Aiming for zero error and precision is the primary goal of the AI solution. Although individuals are sometimes resistant to development, this type of technology will minimise the impediments to access screening and care, thus, mitigating blindness.

Diabetic Retinopathy in the UAE

As indicated by the most recent statistics, the level of diabetes in the UAE has achieved 19.2%. The human eye is viewed as the most powerless organ to be influenced by diabetes. Diabetic retinopathy is a complex phase of diabetes and is separated into two stages. Invasion in the optical layer is the fundamental explanation behind the disintegration of visual impairment. Over the years, therapies for diabetic retinopathy treatment concentrated on controlling side effects and halting the weakening of visual impairment.

However, with the advancement of technologies in eye medical procedure, treatment plans try to restore the patient’s vision to full recovery. With the goal to control the rising frequency of diabetes, the Dubai Diabetes Center (DDC) plans to present AI to distinguish retinopathy, initiate tele-checking of patients who miss their arrangements and present corpulence centres in the Emirate. According to global standards, a patient requires fourteen retinal pictures for cross-examination (Vega, Sanchez-Ante, Falcon-Morales, Sossa, & Guevara, 2015).

The assessed number of analysed diabetics in the UAE surpasses one million. To decipher more than 16 million photographs for each year will require more than 70 eye specialists daily. However, deep learning framework (DLS) can cross-examine diabetic retinopathy and related eye challenges utilising retinal pictures with a high level of precision.

Therefore, AI deployment in the UAE would encourage retinopathy screening many patients, manage resources, cost, and save time. The utilisation of artificial intelligence in the location of diabetic retinopathy can revolutionise screen patterns and improve the quality of healthcare delivery. With this technology, ophthalmologists will need to examine flagged images, unlike screening retina images. Diabetes is a sickness that requires multidisciplinary attention. The coherence of consideration is a fundamental attribute because every patient visits the same expert. Because of the persistent expansion of patients, the number of existing clients has increased, causing delays and expenditures for disappointed customers. Thus, upgrading existing offices with AI systems will enhance healthcare delivery.

Current Challenges (DR Barometer: UAE Report)

Time

For all health care professionals, the average waiting time for an appointment was less than one week (32%), or between one week and a month (32%). For an interview with an ophthalmologist, it was usually between a week and a month for 50% of practices, but for a further 25% of the methods, the wait time was less than one week.

  1. There were long wait times for appointments to see doctors and specialists (27%).
  2. Physicians saw an average of 182 patients per week and 57% (on average) of their patients had diabetes.
  3. Ophthalmologists screened fewer patients each week (74) and (46%) patients had diabetes.
  4. Ophthalmologists reported that the most significant challenges for improving patient outcomes in DED were a late diagnosis (67%, n=2).
  5. On average, 34% of patients seen by ophthalmologists had DR and 24% DME.
  6. The most common waiting time for a screening appointment for DED is less than one week (67%).

Cost

Patients revealed that the cost of care was the primary challenge in controlling diabetes (44%).

Agents and Beneficiaries of AIMS-DRD

Primary healthcare givers are the agents of AIMS-DRD. They conduct daily and routine checks based on patient need. As a result, they provide eye education to patients and encourage them to inform their family member and friends on the need for an eye examination. The prevalence of diabetic retinopathy is caused by inadequate eye education. People do not see the need to conduct a routine eye check, and those who undergo the examinations do not allow the follow-up sessions. Thus, the agents of this program must educate using various sensitisation campaigns.

Related Works

Wong (2017) studied the role of deep learning systems in patient diagnosis. The authors were motivated because the diseases passed the requirement for screening. Diabetic retinopathy is a health challenge that grows into a complex disease. Therefore, the screening and treatment process is considered the safest therapy for patients with the disease. The authors concluded that deep learning systems improved the quality of life of patients suffering from diabetic retinopathy. Vidal-Alaball, Royo, Zapata, Marin-Gomez, and Solans (2019) investigated the algorithm of AI screening. The researchers tested patients with DR. The goal was to build the algorithm for indicators of DR in diabetic patients and approve the screening device for public use. The authors recommended the AI algorithm to improve screening efficiency and accuracy.

Lu et al. (2018) described the improvement and approval of an artificial intelligence-based technology for diabetic retinopathy. The authors deployed a neural system to test images of DR patients. Based on the findings of the study, the researchers suggested that DLA could be utilised with precision in the discovery of eye defects. The innovation improves the effectiveness of DR screening systems.

Kanagasingam et al. (2018) conducted a study to describe the execution of an AI framework for diabetic retinopathy. The results revealed the benefits and difficulties of utilising AI frameworks to diagnose diabetes. The challenges include lower rate of infection, incorrect measurement indicator and poor picture quality. The authors concluded that further assessments of AI frameworks in essential consideration are required.

AI Opportunity in Diabetic Retinopathy

Distinctive screening modalities performed by various professionals will deliver variations in outcome. Clinicians have to screen diabetic retinopathy accurately to prevent permanent blindness. Appropriate education among caregivers is fundamental. The challenges of poor management include poor screening programs, inadequate infrastructures and poor patient history. Artificial intelligence opportunity in diabetic retinopathy empowers specialists to order the retinal pictures successfully. Consequently, to build a diabetic retinopathy-screening framework, successful methods of picture processing must be utilised (Schlegl et al., 2018).

AI enhanced screening and diagnosis in ophthalmology amplifies the specialists’ job. Nevertheless, AI offers the patients innovative opportunities, reducing hindrances to access eye care where an ophthalmologist is not accessible. The technology demonstrates the possibility to assuage the overburdened human services (Khojasteh, Aliahmad, Arjunan, & Kumar, 2018). The AI system supports algorithms that can naturally build a model of complex connections by preparing standardised practices of disease diagnosis (Abràmoff et al., 2016).

The system relies on repeated learning to develop its databases for accurate diagnosis. The machine learning requires a substantial number of retina images in its database (Keel et al., 2017). Similarly, it is pivotal that element determination or extraction requires much experience.

Dubai Health Authority

The annual costs for treating people with type2 diabetes is expected to rise to $563 million (AED 2.07 billion) by 2020 with an additional $89 million

(AED 327 million) spent on type 1. Undiagnosed diabetes will cost $303 million (AED 1.11 billion), and another $84 million (AED 309 million) will be spent treating people with pre-diabetes at $1.04 billion (AED 5.14 billion). The analysis of healthcare costs more than ten years after the implementation of interventions for pre-diabetes and diabetes in the UAE. The gross savings range from $156 million (AED 573 million) to $1.22 billion (AED 4.5 billion) without the cost of implementation.

Using AI and deep learning systems to automate screening has shown tremendous promise. The global healthcare system is overburdened, and the role of AI is seen as the disrupting force to alleviate this problem. The solution will work on deep learning model to detect diabetic retinopathy and other abnormalities in the fundus by initiating a pilot project with the Dubai Health Authority (DHA) to demonstrate the efficacy of this system. Based on the outcome of the pilot project, national expansion can be envisaged, and the hope of making UAE the first nation to offer AI based screening to its citizens can be realised.

Patient Journey: Current Process

Assume a globally standardised workflow, where a patient arrives at the facility to see a clinical assistant who takes the patient for screening. At this stage, a trained operator takes images of the eyes and either render them onto a computer monitor or takes a printout for the ophthalmologist to examine and review the results.

The patient’s journey shows the process for appointment. The stages include registration check, meeting with the ophthalmologist assistance and examination by the ophthalmologist. However, these processes come with its challenges. As shown in figure 2, the patient’s journey could be rescheduled if the physician is busy or the client does not meet the requirement during registration.

The branches of AI include deep learning systems, machine learning, neural network, convolutional neural network (CNN) and supervised learning model (Lee, Taylor, Kalpathy-Cramer, & Tufail, 2017). Each branch of AI has its unique features and similarities. For example, CNN uses retina images as its input. This feature supports the forward function of the screening program. The process as shown in the BPMN diagram shows the stages of the appointment, which start with the first visit. After registration, eye drops are administered to prepare the retina for cross-examination. A nurse who then refers the patient to the physician for medication process then takes the images.

Proposed Business Process Model and Notation (BPMN)

To facilitate the volume of patients expected for screening an automated solution is being proposed by Artelus Software House LLC. As shown in fig. 2, the BPMN displays the flowchart of the business process.

BPMN flowchart.
Fig. 2: BPMN flowchart.

In this context, the patient retina images are evaluated using the deep learning model. Diabetic Retinopathy Detection System (DRDS) is an AI system, which has been designed and developed using deep learning, machine learning and applying AI algorithms to detect the presence of DR in patients during fundus examination screening. This computer-aided diagnostic and detection tool increases the efficiency of the healthcare providers by enabling them to offer higher quality care without overburdening the system (Cios & William Moore, 2002).

Solutions using deep learning equations have a higher rate of accuracy compared with other forms of screening. This software is straightforward to use with a user-friendly interface. It captures and analyses retina images within 15 seconds and prints the report. These reports can also be mailed to patients if required or stored in the local EMR. This sort of technology will be a good scheme in the UAE’s multi-pronged strategy to combat diabetes.

  • Reduce cost. This product will help the government bring down the cost of treating diabetic patients. In the UAE, the current price per person with diabetes is $2,155.90.2
  • Save time. The proposed process saves time for both the patient and physician.
  • Patient benefit. Fringe benefits to Patients by saving their time and cost in consulting doctors will be achieved.

Assuming the same standardised workflow, we are showing how this solution eliminates steps from the workflow thereby accruing benefits to the system. Figure 1 shows the strategy of the proposed technology.

Proposed technology.
Fig. 1: Proposed technology.

The procedure will reduce the role of nurse capturing the images and instead replace them with AI scans. An advantage with this technology is the ability to store the data even for other uses. The image capturing is now automated to AIMS-DRD capture mode, which is an application of machine learning. The proposed model shows an automated screening pattern. The patient visits the vision centre and profiles for the AIM-DRD capture. The consult ophthalmologist will review the image analysis.

Software Process Modeling

There are many deep learning models used for image capture. For the proposed solutions, this paper would discuss machine learning and neural networks.

Machine Learning

Machine learning deals with the development and investigation of frameworks that process information from databases. The algorithms of this procedure utilise computational strategy to handle data without depending on a foreordained equation (Yun et al., 2008).

The algorithms adaptively improve their execution as the number of tests accessible for processing increases. The assumptions of machine learning state that the program information based on experience E from some tasks T and execution measure P, if its execution of assignments in T as estimated by P, improves with experience E. The focus of machine learning manages representation and speculation. Representing the information cases and capacities assessed on these examples are components of the machine-learning framework (Krizhevsky, Sutskever, & Hinton, 2017).

Machine learning power is the ability of the AI system to perform effectively in new and inconspicuous information activities having processed the knowledge database. Data processing precedents originate for prediction and the specialist must design a general model that empowers the system to create accurate forecast (Takahashi, Tampo, Arai, Inoue, & Kawashima, 2017). The efficiency of machine learning speculation is assessed based on the capacity to repeat the process accurately (Vayena, Blasimme, & Cohen, 2018). Two fundamental types of machine learning include supervised and unsupervised learning models.

Supervised Learning Model

As the name implies, it is the task of construing a capacity for supervised processing information. Processing information for administered learning incorporates many models with combined data subjects and estimated yield (Krause et al., 2018). A monitored process evaluates the processing data and produces a derived capacity, which is called the regressed function (Gupta & Chhikara, 2018). By implication, the regressed function should decode the right yield for any input image.

This requires the learning system to speculate from the processing information to unknown circumstances in a fixed pattern. A straightforward similarity to the supervised model is the connection between a student and an instructor. The instructor discusses a topic for a specific discipline. The instructor provides a different set of questions and solutions to areas of concentration. Finally, the instructor sets a test paper to evaluate the student’s performance.

The SLM acquires information from images and processed output. When the processing data is completed, new photographs are uploaded into the model, and the framework creates the yield utilising knowledge it picked from its terminal. The supervised learning system operates with this function.

Algorithms: Neural Networks

The neural network is a subclass of algorithms that manage different model in AI. Many machine-learning models are striking for being versatile. Each hub of the neural system has its circle of information about guidelines and functionalities to create its algorithms through encounters gained from past procedures that do not depend on neural networks (Tarassenko & Watkinson, 2018). Neural systems are appropriate for distinguishing non-direct designs, as in examples where there is no connection between the information and yield.

Versatile loads describe neural systems along channels between neurons that can be altered by a learning equation created from observed processes and database. The system uses an appropriate cost function to evaluate the input algorithms (Crossland et al., 2016). Concisely, it can modify itself to the changing conditions as it gains from the learning field and unknown data terminals.

Convolutional Neural Network (CNN)

CNN is a modelled algorithm that collects images, assigns weights to each input and have the capacity to separate one input from the other (Alaskar, Hussain, Al-Aseem, Liatsis, & Al-Jumeily, 2019). The startup phase of a CNN model is much lower when contrasted with different algorithms. A CNN model can process image attributes and characteristics. The design of a ConvNet is similar to that of the human mind and invigorated by the association of the visual cortex (Xu, Feng, & Mi, 2017).

The neuron system reacts to improvements in a confined area called the receptive field. An accumulation of such fields covers the optical zone. In situations of double images, the strategy may demonstrate an average score while performing a forecast of classes. The technique will have little effect on complex models. Thus, CNN can capture the conditions in a picture through the application of significant filters (Tsao, Chan, & Su, 2018).

The design allows a superior fitting to the picture dataset because of the decrease in the number of values and reusability of loads. As a result, the system can be prepared to comprehend the advancement of images within the visual zone. The system categorises pictures into a structure, which is easy to process without losing attributes vital for accurate evaluation. This property is important to designers who create effective systems in learning and scalable to many images. The goal of the operation is to extricate specific features from the information picture.

The screening of the DR has attracted many considerations with studies, experimenting scale-aneurysm, exudation and neo-vascularisation discovery. Most of these studies utilise the fundus pictures as the sampled data. The system collects images named with symptomatic lesions, separate their qualities and assemble a model. Based on the constructed model, the system can distinguish the new input and predict an accurate output. Imani, Pourreza and Banaee (2015) structured a procedure to identify vein vessels using a morphological analysis.

The cycle of model construction as shown in figure 3, starts with data cleansing and processing. The next phase begins with model training using a machine-learning algorithm. Once the constructed model is completed, the system can predict DR variations considering the result of accuracy.

CNN Model.
Fig. 3: CNN Model.

Software Process Model: Agile Approach and Reinforcement Learning

The agile approach assist developers respond to uncertainty. Reinforcement learning (RL) is the teaching of AI models to predict informed outputs. The specialist studies how to accomplish an objective in an unsure complex condition (Jackson, Yaqub, & Li, 2019). In RL, artificial intelligence faces a diversion-like circumstance. The automated system utilises experimentation to concoct an answer to the challenge. The system is designed to collect rewards or punishment for the activities it performs. It will probably amplify the reward.

Although the specialist sets the reward function, they give the model no indication or recommendation on how to understand each task. It depends on the model to determine how to accumulate the reward, beginning from the preliminaries and completing with advanced strategies and superhuman aptitudes (Jackson et al., 2019). By utilising the intensity of repeated trials, RL is an effective method to indicate machine inventiveness. Thus, AI can accumulate knowledge from parallel interactive tasks if the RL algorithm operates on strong system infrastructure.

Dataset and Pre-Processing Consideration

When configuring the AI algorithm, diverse emphases and training will be directed on the equivalent dataset. Before beginning each new configuration, the convolutional neural network will be utilised randomly. The data set will be divided into two sets, comprising 80% of the pictures for the pre-processing phase and 20% for the approval stage. A data set of 80% will be sufficient to prevent alteration in parameter estimation. Utilising the remaining 20% for cross-approval will be enough to maintain a strategic distance in the measurement metric. Based on the outcome of the first test (pre-processing and approval), the system will modify the ratio of 80/20.

After each configuring process, the estimated values of precision and loss for each phase will be documented. The system will produce a graphical representation of estimations of exactness and misfortune, both for the pre-processing dataset and the approval dataset (Jackson et al., 2019). With these diagrams, the architect will extricate useful data to know the numbers of successful training sessions, regardless of whether the learning rate is sufficient. With the approval dataset, the designer will compute true positives (TPs), true negatives (TN), false positives (FPs) and false negatives (FNs). Using a picture-by-picture design can be used to rename and evaluate the specimen.

During the proposed patient journey, the assigned physicians will assess the fundus pictures and report their discoveries in the electronic clinical notes. The report will be collated and tagged with patient identification. OPT retina will be used to load the dataset on a web application. The pictures will become accessible for the AI prediction and human specialists to revalidate and characterise the investigation (Jackson et al., 2019). A blinded twofold appraisal by the retina experts will reduce errors or contradictions. Once the study is finalised, the findings will be renamed using verifiable identification to ascertain the execution measurements for the examinations. The experts will quantify the execution of the AI model utilising the open Messidor-2 dataset.

  • Sensitivity or genuine positive rate=TP/ (TP+FN)
  • Specificity or genuine negative rate=TN/ (TN+FP)
  • Accuracy= (TP+TN) / (TP+FP+FN+TN)
  • Where ACU = Area under the collector curve.

Table 1: Training Algorithm.

Remark Test values
Bad test 0.5, 0.6
Regular test 0.6, 0.75
Good test 0.75, 0.9
Very good test 0.9, 0.97
Excellent test 0.97, 1

Sources of Error or Bias Arising from the Expected Data Sources

The challenges of failure resulting from the predicted data sources include data volume, handling capacity, external interference, image quality, mismatched tags and class distribution.

Data Volume

It is difficult to exchange and store gigabytes of pictures. For this challenge, the research would enlist Amazon Web Ltd to acquire the transmission capacity and storage ability to protect the enormous volume of specimen data.

Image Processing Capacity

The iterative process of deep learning systems forces huge expense, cost, time and resource infrastructure. The system will be designed to match images with low resolutions to mitigate image-processing capacity.

External Interference

Specimen errors include mismatched images, poor image handling, poor image quality, and class distribution. This challenge can be prevented with accurate recording skill and job efficiency.

Proposed Machine Learning Model: Convolutional Neural Network (CNN)

CNN is a modelled algorithm that collects images, assigns weights to each input and have the capacity to separate one input from the other. The startup phase of a CNN model is much lower when contrasted with different algorithms. A CNN model can process image attributes and characteristics. The design of a ConvNet is similar to that of the human mind and invigorated by the association of the visual cortex.

The neuron system reacts to improvements in a confined area called the receptive field. An accumulation of such fields covers the optical zone. In situations of double images, the strategy may demonstrate an average score while performing a forecast of classes (Akyol, Sen, & Bayır, 2016). CNN captures the conditions in a picture through the application of significant filters (Lakhani & Sundaram, 2017). The design allows a superior fitting to the picture dataset because of the decrease in the number of values and reusability of loads.

As a result, the system can be prepared to comprehend the advancement of images within the visual zone. The job of the system is to categorise pictures into a structure, which is easy to process, without losing attributes vital for accurate evaluation. This property is important to designers who create effective systems in learning and scalable to many images. The goal of the operation is to extricate specific features from the information picture (Yang et al., 2018). The screening of the DR has attracted many considerations with studies, experimenting scale-aneurysm, exudation and neo-vascularisation discovery. Most of these studies utilise the fundus pictures as the sampled data (Mansour, 2018).

The system collects images named with symptomatic lesions, separate their qualities and assemble a model. Based on the constructed model, the system can distinguish the new input and predict an accurate output.

The neural network is a subclass of algorithms that manage different model in AI. Many machine-learning models are striking for being versatile. Each hub of the neural system has its circle of information about guidelines and functionalities to create its algorithms through encounters gained from past procedures that do not depend on neural networks (Ting et al., 2017). Neural systems are appropriate for distinguishing non-direct designs, as in examples where there is no connection between the information and yield.

Versatile loads describe neural systems along channels between neurons that can be altered by a learning equation created from observed processes and database. The system uses an appropriate cost function to evaluate the input algorithms. Concisely, it can adjust itself to the changing conditions as it gains from the learning field and unknown data terminals.

Ethical Social and Security Implications of AI

The moral, social and security issues linked with AI is influenced by information; mechanisation, innovation and the utilisation of assistive technology. The challenges of AI include administrative concerns, confidentiality, data privacy, unethical algorithm and legal liability. Reviewing data awareness training for all system users, periodic security reminders and user education concerning virus protection are some concerns of the AI clients. AI system takes decisions based on the training set through a learning model. Sometimes, robots make a wrong decision leading to fatal outcome or interruption of service. Such scenarios raise the question of whom to hold liable, given that it is almost impossible to identify how the system has arrived at its decision.

Safety

Safety issues include hardware control, treatment and complications due to false judgment. Error in opinion of the system could cause permanent blindness.

Transparency

It is difficult to ascertain the rationale that produces the yields created by AI. The mechanism of artificial intelligence is restrictive and purposely hidden, although some models are complex for human comprehension. The arcane knowledge for approving the yields of AI frameworks creates ethical, social and security challenges in healthcare management.

Information Bias and Equity

Although artificial intelligence can diminish human prejudice and poor judgment, they can likewise reflect and strengthen predispositions in the information used to train them. Issues have been discussed about the capability of artificial intelligence to cause discrimination because of its hidden abilities. The perception could be sexual orientation, ethnicity, image and age. Based on this context, bias attributes could be inserted during the pre-processing phase.

Impact on Patients

Artificial intelligence applications assist patients in testing vitals and simple body biometrics. AI frameworks could improve recovery, freedom and empower patients who have been admitted in care facilities. However, concerns have been raised about losing human contact and expanded social segregation if AI advancements are utilised to supplant staff or family time with patients. Artificial intelligence frameworks could negatively affect self-sufficiency when they confine decisions dependent on computations.

Consequently, when deep learning systems are utilised to make predictions or treatment plan, and a physician cannot clarify how these recommendations were reached; this could be viewed as confining the patient’s right to make informed decisions about their health. Applications that mirror human reasoning raise the likelihood that the client cannot judge whether they are speaking with a genuine personality or technology. The technology could be termed fraudulent and deceptive.

Impact on Healthcare Givers

Clinical professionals may feel compromised if artificial intelligence systems test their ability. Consequently, the moral commitments of human service experts towards patients may be influenced by the utilisation of artificial intelligence infrastructures. Based on this context, health professionals may feel insecure or attached to computed aided systems to the detriment of their cognitive thinking.

Information Privacy and Security

AI systems utilise data that are considered confidential. The information could be spammed, hacked, or stolen by unauthorised users. Such uncertainty affects the acceptability of AI systems.

Malignant use of AI

While AI can be utilised to improve the quality of human lives, it could be used for destructive purposes. AI could be employed for clandestine observation or screening. AI could be utilised to carry out digital attacks at a lower expense and higher scale. This challenge has prompted calls to address the double use nature of AI and plan for conceivable vindictive employments of AI technologies.

Conclusion

AI frameworks assist doctors by giving current restorative data from diaries, course readings and clinical practices to advise appropriate patient care. Consequently, the AI framework can decrease unavoidable demonstrative and remedial blunders in clinical exercise. Given the novelty of this solution, it is important to understand the difficulties associated with its deployment. The aim is to collaborate with DHA to reduce the cost of implementation.

If the recommendations were adopted, the facility would improve the screening of DR by saving time, cost of treatment, prevent blindness and enhance the quality of healthcare delivery. The capacity to represent the human personality and conduct complex evaluation is named artificial intelligence. Given the difficulties and unpredictability related to AI, numerous analysts directed their concentration toward narrow AI, which is the ability to conduct specific assignments.

AI procedures have sent huge waves over healthcare insurance, fuelling a functioning view of whether AI specialists will displace human doctors. Although machines cannot replace human doctors, AI can help doctors settle on better clinical choices or even supplant human judgment in certain useful territories such as radiology. The expanding accessibility of human service information and quick improvement of big data strategies supports the ongoing applications of artificial intelligence in healthcare services.

References

Abràmoff, M. D., Lou, Y., Erginay, A., Clarida, W., Amelon, R., Folk, J. C., & Niemeijer, M. (2016). . Investigative Opthalmology & Visual Science, 57(13), 5200. Web.

Akyol, K., Şen, B., & Bayır, Ş. (2016). . Computational and Mathematical Methods in Medicine, 1–10. Web.

Alaskar, H., Hussain, A., Al-Aseem, N., Liatsis, P., & Al-Jumeily, D. (2019). . Sensors, 19(6), 1265. Web.

Cios, K. J., & William Moore, G. (2002). . Artificial Intelligence in Medicine, 26(1–2), 1–24. Web.

Crossland, L., Askew, D., Ware, R., Cranstoun, P., Mitchell, P., Bryett, A., & Jackson, C. (2016). . Journal of Diabetes Research, 1–7. Web.

Gupta, A., & Chhikara, R. (2018). . Procedia Computer Science, 132, 1432–1440. Web.

Imani, E., Pourreza, H.-R., & Banaee, T. (2015). . Computerized Medical Imaging and Graphics, 43, 78–88. Web.

Jackson, S., Yaqub, M., & Li, C.-X. (2019). . Frontiers in Big Data, 1. Web.

Kanagasingam, Y., Xiao, D., Vignarajan, J., Preetham, A., Tay-Kearney, M.-L., & Mehrotra, A. (2018). . JAMA Network Open, 1(5), e182665. Web.

Keel, S., Xie, J., Foreman, J., van Wijngaarden, P., Taylor, H. R., & Dirani, M. (2017). The prevalence of diabetic retinopathy in Australian adults with self-reported diabetes. Ophthalmology, 124(7), 977–984. Web.

Khojasteh, P., Aliahmad, B., Arjunan, S. P., & Kumar, D. K. (2018). . In 2018 40th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) (pp. 5938–5941). Honolulu, HI: IEEE. Web.

Krause, J., Gulshan, V., Rahimy, E., Karth, P., Widner, K., Corrado, G. S., … Webster, D. R. (2018). . Ophthalmology, 125(8), 1264–1272. Web.

Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017). . Communications of the ACM, 60(6), 84–90. Web.

Lakhani, P., & Sundaram, B. (2017). . Radiology, 284(2), 574–582. Web.

Lee, A., Taylor, P., Kalpathy-Cramer, J., & Tufail, A. (2017). Ophthalmology, 124(12), 1726–1728. Web.

Lu, W., Tong, Y., Yu, Y., Xing, Y., Chen, C., & Shen, Y. (2018). . Journal of Ophthalmology, 1–15. Web.

Mansour, R. F. (2018). . Biomedical Engineering Letters, 8(1), 41–57. Web.

Mirsharif, Q., Tajeripour, F., & Pourreza, H. (2013). . Computerized Medical Imaging and Graphics, 37(7–8), 607–617. Web.

Niemeijer, M., Abràmoff, M. D., & van Ginneken, B. (2009). . Medical Image Analysis, 13(6), 859–870. Web.

Raja, C., & Gangatharan, N. (2015). . Computers in Biology and Medicine, 63, 196–207. Web.

Schlegl, T., Waldstein, S. M., Bogunovic, H., Endstraßer, F., Sadeghipour, A., Philip, A.-M., … Schmidt-Erfurth, U. (2018). . Ophthalmology, 125(4), 549–558. Web.

Takahashi, H., Tampo, H., Arai, Y., Inoue, Y., & Kawashima, H. (2017a). . PLOS ONE, 12(6), e0179790. Web.

Tarassenko, L., & Watkinson, P. (2018). . The Lancet, 391(10127), 1260. Web.

Ting, D. S. W., Cheung, C. Y.-L., Lim, G., Tan, G. S. W., Quang, N. D., Gan, A., … Wong, T. Y. (2017). . JAMA, 318(22), 2211. Web.

Tsao, H.-Y., Chan, P.-Y., & Su, E. C.-Y. (2018a). . BMC Bioinformatics, 19(S9). Web.

Vayena, E., Blasimme, A., & Cohen, I. G. (2018). . PLOS Medicine, 15(11), e1002689. Web.

Vega, R., Sanchez-Ante, G., Falcon-Morales, L. E., Sossa, H., & Guevara, E. (2015). . Computers in Biology and Medicine, 58, 20–30. Web.

Vidal-Alaball, J., Royo Fibla, D., Zapata, M. A., Marin-Gomez, F. X., & Solans Fernandez, O. (2019). . JMIR Research Protocols, 8(2), e12539. Web.

Weng, S. F., Reps, J., Kai, J., Garibaldi, J. M., & Qureshi, N. (2017). PLOS ONE, 12(4), e0174944. Web.

Xu, K., Feng, D., & Mi, H. (2017). . Molecules, 22(12), 2054. Web.

Yang, Y., Yan, L.-F., Zhang, X., Han, Y., Nan, H.-Y., Hu, Y.-C., … Wang, W. (2018). . Frontiers in Neuroscience, 12. Web.

Yazid, H., Arof, H., & Isa, H. M. (2012). . Journal of Medical Systems, 36(3), 1997–2004. Web.

Yun, W. L., Rajendra Acharya, U., Venkatesh, Y. V., Chee, C., Min, L. C., & Ng, E. Y. K. (2008). . Information Sciences, 178(1), 106–121. Web.

This report on Artificial Intelligence in Healthcare Delivery was written and submitted by your fellow student. You are free to use it for research and reference purposes in order to write your own paper; however, you must cite it accordingly.
Removal Request
If you are the copyright owner of this paper and no longer wish to have your work published on IvyPanda.
Request the removal

Need a custom Report sample written from scratch by
professional specifically for you?

Writer online avatar
Writer online avatar
Writer online avatar
Writer online avatar
Writer online avatar
Writer online avatar
Writer online avatar
Writer online avatar
Writer online avatar
Writer online avatar
Writer online avatar
Writer online avatar

certified writers online

Cite This paper
Select a referencing style:

Reference

IvyPanda. (2020, December 3). Artificial Intelligence in Healthcare Delivery. Retrieved from https://ivypanda.com/essays/artificial-intelligence-in-healthcare-delivery/

Work Cited

"Artificial Intelligence in Healthcare Delivery." IvyPanda, 3 Dec. 2020, ivypanda.com/essays/artificial-intelligence-in-healthcare-delivery/.

1. IvyPanda. "Artificial Intelligence in Healthcare Delivery." December 3, 2020. https://ivypanda.com/essays/artificial-intelligence-in-healthcare-delivery/.


Bibliography


IvyPanda. "Artificial Intelligence in Healthcare Delivery." December 3, 2020. https://ivypanda.com/essays/artificial-intelligence-in-healthcare-delivery/.

References

IvyPanda. 2020. "Artificial Intelligence in Healthcare Delivery." December 3, 2020. https://ivypanda.com/essays/artificial-intelligence-in-healthcare-delivery/.

References

IvyPanda. (2020) 'Artificial Intelligence in Healthcare Delivery'. 3 December.

More related papers
Psst... Stuck with your
assignment? 😱
Hellen
Online
Psst... Stuck with your assignment? 😱
Do you need an essay to be done?
What type of assignment 📝 do you need?
How many pages (words) do you need? Let's see if we can help you!