The Opportunities of Explainable AI in Healthcare

Aftab Ara
Assistant Professor, Business Department, University of Hail,Saudi Arabia
Automatic Summary

AI in Healthcare: Challenges and Opportunities

It's an honor to introduce Dr. Asta, an Assistant Professor at the Department of Business, University of High Sodi Arabia, who recently presented at the Tech Global Conference 2023. Her insights into the opportunities and challenges of incorporating AI into the healthcare sector shine a light on the evolutionary journey of today's healthcare industry.

Impact of AI in Healthcare

Artificial Intelligence (AI) is changing our lives in many ways, especially within the healthcare sector. Dr. Asta explored the different dimensions of AI's role in healthcare. The areas impacted by AI in healthcare range from reducing labor costs and automating processes to enhancing the customer experience, aiding in treatment, improving medical diagnosis, and accelerating drug discovery.

  • Cost Reduction: AI automation has not only improved product quality but also conspicuously reduced labor-intensive processes, implying a considerable reduction in labor cost.
  • Customer Engagement: In a healthcare context, AI's ability to generate personalized recommendations and information greatly enhances customer experience, making processes more engaging and seamless.
  • Treatment Aid: AI helps detect diseases faster and provides personalized treatment plans, contributing concretely to patient care. Its ability to automate drug discovery and diagnosis processes is revolutionizing the healthcare landscape.
  • Medical Diagnosis and Drug Discovery: Deep learning algorithms in software can analyze medical images to diagnose various diseases. AI has also accelerated drug discovery, making the process faster and cheaper.

AI Challenges in Healthcare

Despite numerous benefits, AI application in healthcare comes with specific challenges. Dr. Asta highlighted significant issues like increased automation leading to job loss, privacy concerns, ethical challenges, and potential risks concerning cyberattacks.

  • Job loss: Owing to AI technology advancements, certain roles, particularly in imaging and diagnostics, are at risk.
  • Privacy Concerns: AI usage raises significant concerns about data usage and protection. It's crucial that organizations safeguard their data assets within high-security, HIPAA compliant systems.
  • Ethical Challenges: These concerns revolve around bias, discrimination, and accountability. Potential solutions could include implementing ethical guidelines and standards, promoting diversity, and ensuring ongoing human oversight.

An alarming issue is the risk of cyberattacks which can manipulate AI models, leading to incorrect diagnosis. This poses a significant risk to patient safety and emphasizes the need for robust systems.

WHO's Recommendations For AI in Healthcare

In 2021, the World Health Organization (WHO) released the first global report on ethics and governance of AI in healthcare. It emphasized the prospective health disparity that can emerge due to AI systems trained on data collected mostly from patients in high-income settings.

WHO's recommendations are centered around protecting human autonomy, human well-being, ensuring transparency, fostering responsibility and accountability, and ensuring AI systems are inclusive, equitable, responsive, and sustainable.

The Future of AI in Healthcare

Even though the times are exciting yet filled with challenges for AI in healthcare, it is crucial to tackle these find solutions for a sustainable future. As humans, we have the opportunity to shape the patient care's future by overcoming fear and hesitation of automation, understanding AI's ethical concerns, ensuring data security, and achieving a return on investment.

As Dr. Asta concluded, it is the perfect time to reshape the healthcare ecosystem, make intelligent use of AI, and chart a course towards a more sustainable healthcare system.

Thank you, Dr. Asta, for enlightening us about AI's capabilities, implications, and future in healthcare, and we wish all conference participants the best of luck.

Video Transcription

Explain I in health care. Uh Thanks a lot for accepting my presentation in the Tech Global Conference 2023. I'm Doctor Asta uh working as an assistant professor in the Department of Business University of High Sodi Arabia.Today, we know that there are many challenges of which are in healthcare sector. Um My key takeaways in the session are the opportunities what A I has brought to the healthcare sector and to also go through the challenges which are there for the A I application and the healthcare. Presently, we know that with the introduction of A I in health care, there are several reports which are produced um in many countries. And uh that has led to quite a lot of ethical challenges relating to the data representation A I algorithm which are untrained sometimes. And though A I has a lot of impact in our lives, which I will have a go through here. So it has impacted in our life in red, reducing the cost in uh consumer engagement in helping in the treatment also in uh making decisions for uh clinical cases. Uh It has also helped in um new drug deal recoveries. And by doing new medical diagnosis as well as we are hearing of uh robotic surgery, which are going on uh from different places and there are worse nursing assistance.

Um also fraud detection clinical trials which are going on. So, these are the many ways in which uh A I has um impacted our life. Uh One of the way in which A A I has impacted our life is uh it has uh reduced cost. Uh labor intensive processes has been uh quite automated by A I nowadays. So the labor cost has reduced. Uh In addition, it has also improved in uh quality of the products. And moreover, as A I technology has advanced healthcare organizations are using it in a very large and complex operations. So this has helped us in understanding how they are making decisions. Thus allowing the healthcare professionals to focus their time in giving a very high quality and high value activities. So this has also helped employees to focus on greater and more complex tasks. So, and the implementation of robotic process automation reduction, operating uh cost reduction uh and maximizing the valuable resources. So as a result of increasing availability of A I solutions A I healthcare organizations have the opportunity to leverage them for improving their efficiency and as well as uh saving their costs.

Um It has also A I has also impacted in enhancing the computer ex uh customer experience by providing more customers with personalized recommendations. And informations. Uh One can also make the process quite engaging with the company and make it more seamless. Uh in the healthcare context.

We ca we can say that it has been providing people um and the patients with more information about their care and how they can be treated. So it has been um helping the patients in aiding in treatment by discovering um disease by detecting diseases in the faster way and also giving quite personal treatment, personalized treatment plans and also automating um certain processes like discovery of drugs and diagnosis as well.

A I has also helped in um typical employee computerized predictive analytics algorithms which can filter organize and search for patterns in uh big data. So this helps uh from multiple sources by getting the data. It gives uh uh probability analysis by which the health care providers can make very fast and informed decisions. So hospitals and healthcare solutions are introducing these platforms to enable the use of machine learning to assist with diagnostic disease uh decisions and also for predicting the treatment outcomes.

A I has also um helped in medical diagnosis. There are software with deep learning algorithms that can analyze medical images to diagnose uh cardiovascular disease, wrist fractures, strokes, diabetic retinopathy as well. There are also other applications undergoing clinical trials uh including programs which can be used to dry growths, other kinds of uh breast and skin diseases, congenital cataract diseases, et cetera. Um A I has also improved the dis drug discovery and it has made them faster and cheaper.

So by predicting how uh potential drugs can have uh their impact in the body. And by discarding the dead end components before they leave the computer, the mission e learning models uh helps in cutting down on the need for a lot of lab work. So, uh in future, we can see that the drugs are um manufactured and they are discovered in 1/10 of the time from what it is done now. So um it is helpful in life changing and it is a game changing. There are game changing drugs on a scale and at a pace, it can improve as we have never seen this before. And as I told previously that A I is also helping for uh drug uh for robotic surgery. And um it it has improved and it has been introduced into surgery just most recently. And there is a strong root in imaging and navigation of the techniques focused on feature detection and computerassisted interventions for both the pre operative planning and the intraoperative guidance.

So A I is gradually changing the practice of surgery with technological advances in imaging, navigation and robotic interventions. We are also getting now uh A I virtual nursing assistance uh which can help in the fo follow up with treatments, giving clinical advice uh between the doctor visits.

And we can see that there are a lot of uh examples in uh reducing in cost in consume consummation scale. And it helps in engaging in screening, assessing and supporting larger populations. A I has also been helpful in medical risk prevention. So um it helps to evaluate the unstructured data about risk behavior in the organization's operations. And these algorithms can identify the pattern of behavior related to what has happened in the past incident and then they are risk predictors. So this also helps in the uh A I being able to identify and track whatever fraudulent activities are in health care, whatever irregularities are there. Um It is also helpful like I told fraud detection where you you can uh detect the scams in the healthcare. This technology has a good potential to help in reducing the fraud and in protecting the patients from any kinds of uh fraudulent practices. So A I is quite effective in this field though it's in the early stages, there are great potential for it in the future. Uh There are also clinical trials which are done with the help of uh A I. So this has a reduced uh uh testing on uh human beings. So it has streamlined operation process innovation, innovated the trial design and also has accelerated the uh data collection. So there are also challenges ahead for A I applications in healthcare like A I investment in the health care.

According to the survey, it was found that 85% of the respondents expected to increase uh their investment in A I in the next year compared to what 73% did in the previous study, then there is also uh a risk uh due to the A I in healthcare. So in light of um A is uh potential risk in the cybersecurity, uh there is uh chances that uh A is decision and lack of transparency. Healthcare organizations, they have to um establish appropriate governance and oversight of algorithms and data by implementing and the initiatives so that they can tackle the A I bias in a very uh trustworthy manner. Uh With those challenges, there are also challenges of uh job loss.

Um The development of these A I technologies uh do outperform humans in the news all the time we are seeing this and this trend is only expected to continue further. The field of uh image analytics and diagnostics is where many of the significant advances are made. So uh making radiologists and pathologists particularly quite um susceptible. So in the recent years, it has been found that A I BASED imaging solutions have changed from like academic research to they have become more uh commercial. So instruments are uh quite available for um wide range of eye as well as the skin conditions as well as for detecting cancer. And um they are also supporting in the measurements required for clinical diagnosis um as we are seeing in the research reports, mm there are some these symptoms uh which helps in um tiresome jobs and they do compete with some diagnostic skills of the experienced pathologist as well.

And the radiologist too, like counting the number of cells which is dividing in the cancer tissues, et cetera, the employment of automated systems. Uh however, do present some ethical um questions in other fields as well. Uh A I privacy is also one of the challenge A I privacy is a topic which concerns how patients medical data is used and how it is protected by the A I technology according to uh protecting to an individual's information and privacy becomes even more important as with the exchange of medical information between patients, physicians and the care teams uh which are there.

So we have seen that most of the organizations um are advised uh that they should keep their data assets quite closely guarded in very high secure and in H IP P A compliance systems. Um in light of the epidemic of ransomware knockout punches from cyber attacks, there are very much chances and um CIS Os have also every right to be reluctant to lower their job, job bridges and allow data to move freely so that um they can move in and out of the data. Um Then uh in 2021 as we can see, there has been a study which was done in Pittsburgh Research, uh which is also published in nature Communication. And here it was found that cyberattacks um employing uh fabricated medical images which deceives A I models uh to more conventional cyber attacks. And also the patients have privacy issues here. So A I models can be tricked by hacking or by fabricating the medical uh photos as well.

So as uh we have seen and this is quite a grave issue that there was a study which was done on the attack in which the uh malicious actors tried to, tried to manipulate the photos of other data pieces to lead all the A I network to make a wrong inference. So here uh what happened, the researchers trained the deep learning algorithm uh to distinguish between the malignant and the benign cases and the accuracy rate was like 80%. So uh so to confuse the model, the researchers created a generative um adverse network also. And a computer software that created false image by erasing the malignant patches from other positive or negative images. Then the model was deceived by 69.1% of the false images. So, out of the 44 positive images which was made to look negative. The model identified only 42 as negative.

So out of 390 negative images to look positive A A I model classified 209 as um positive. So uh what we can uh know from this study is that this type of attack is possible and it could lead to A I models to make wrong diagnosis. So this this is a very big issue for the patient's safety. As W uh um Shan Wu says that this uh studies senior author and he's an associate professor of radiology, biomedical informatics and bioengineering in the University of Pittsburgh. So he explained this in a conference release how by undergoing the A I models behave under adverse attacks and in medical context. And we can start thinking to make uh models which should be safe and which are more robust. So um uh there are also ethical challenges due to A I, uh A I can be used to improve diagnosis, they can be used for treatment, they can be used for care. But it also has some ethical challenges. Like there are bias, there is discrimination, there is accountability and transparency. Some possible solutions could include like having good quality data by having diversity by having ethical guidelines and standards. So uh which can involve the other stakeholders and patients in making decisions so that the human oversight and explain is there. So we have to identify and address the biases. Quite early. Developers, developers should know and they should introduce bias to A I ALGO algorithms.

They have to train the algorithms. And uh in 19, in 2021 the who also released the first global report um on the ethics and governance of A I in health care. According to who uh who emphasized the potential uh health disparity that can emerge as a result of A I, particularly due to A I systems which are trained on data collection from patients in very high care income setting. So according to who it recommends to protect the human autonomy, it uh tries to pro promote the human being are safe and in public interest, ensure transparency, explainability and intelligently. And it also fosters responsibility and accountability, uh ensuring inclusiveness as well as equity and promoting that A I is responsive and sustainable. So A I and the future of work, there are challenges in healthcare to take care of, but we have to overcome the hesitation and fear of automation. There should be an understanding about A is um ethical concern so that they can combat the device and there should be data security and there has to be a delivery of return of investment and um there should be proper dep deployment in scale.

Uh As I have already gone through, this is the whole report about uh how it wants to protect, promote, ensure, foster, ensure and promote that A I is responsible for a more sustainable world. So here, um though it's very exciting, confusing and frustrating time for the health care, but um it is important and it is a very mature time for A I that it should not only add to the mixed emotion of what's going on, but there should be um clear answers to all these challenges at the moment because we as humans still have the opportunity to take um the control in our hands and make hard choices for uh shaping the future of the patient care.

So I would like to say that the healthcare ecosystem has to start somewhere from the scratch though, even if it is a good time and the best and the right time to start. So, thank you so much for watching. And uh I hope uh that good luck to all the other participants who are in the conference, who are speaking at the conference.