MRI Scans and AI: A Novel Approach to Diagnose Parkinson’s Disease in a Matter of Seconds
How AI is helping the healthcare industry
By Shehraan Hafiz
Introduction to AI
Artificial intelligence (AI) is the use of computers and machines to simulate human intelligence and perform complex tasks. AI has vast potential to improve healthcare in various ways, from diagnosis and treatment to research and administration. In this review, we will explore some of the current and future applications of AI in healthcare, as well as the benefits and challenges of this technology.
Why We Need Better Healthcare
Healthcare is one of the most important and challenging domains for human society, as it involves the well-being, survival, and quality of life of millions of people. Healthcare also faces many problems, such as increasing demand, rising costs, limited resources, human errors, and variability in quality and outcomes. Therefore, there is a need for innovative solutions that can enhance the efficiency, effectiveness, and accessibility of healthcare services.
AI’s Potential to Assist our Healthcare System
AI in disease diagnosis and treatment is one of the most promising applications of AI in healthcare. Diagnosis is the process of identifying and characterizing a disease or condition based on the signs and symptoms of a patient. Treatment is the process of providing medical interventions to cure or alleviate a disease or condition. Both diagnosis and treatment require a high level of expertise, experience, and judgment from clinicians, as well as the integration and analysis of large amounts of data, such as medical records, images, and test results.
AI can also advance scientific knowledge and innovation in healthcare by facilitating the generation, dissemination, and application of new insights and discoveries. Healthcare research is the systematic investigation and exploration of the phenomena and problems related to health and disease, such as the causes, mechanisms, diagnosis, treatment, prevention, and outcomes of diseases. Healthcare research is crucial for improving the understanding, evidence, and practice of healthcare, but it also involves many complex and challenging tasks that require a lot of data, skills, and resources.
The Applicability of AI
An AI is highly efficient as all that an AI model requires to run is code and a machine to train it on. This training itself can take anywhere from a few hours to many months. However, this is nothing in comparison to the 10 years of studying that an average doctor is required to do, to be able to treat patients. And that is only taking into account the post-secondary education — doctors, like many other professionals, also have to complete 14 years of grade school prior to beginning their focus on the healthcare field. This level of sophisticated training and education not only wastes time, it also wastes money— money which could otherwise be dedicated towards creating servers and advancing AI capabilities.
Overall, AI’s use in healthcare offers many benefits, such as improving the quality, accessibility, and affordability of care, enhancing the productivity, satisfaction, and safety of healthcare professionals and patients, and accelerating scientific progress and innovation in healthcare. However, AI in healthcare also poses many challenges, such as ensuring the accuracy, reliability, and validity of AI systems, protecting the privacy, security, and consent of patients and healthcare data, addressing the ethical, legal, and social implications of AI in healthcare, and ensuring the human oversight, accountability, and responsibility of AI in healthcare.
AI companies currently operating in MedTech
Aidence: Amsterdam-based Clinician-oriented company, which provides deep learning systems for lung cancer diagnosis that include clear markers for nodules.
Aiva Health: A Los Angeles based company which is the first to provide voice-powered care assistant that connects patients with the right physicians. It is most prominently used in hospital patient rooms and senior homes.
Bot MD: A Signaporean company that provides a clinician-oriented bot assistant that can answer clinical question, transcribe dictated case notes, as well as automatically organize files and images.
eMed (previously known as “Babylon Health”): A largescale virtual healthcare provider that uses Natural Language Processing to create an internationally accessible virtual healthcare provider. It is able to use its knowledge + patient data to provide information on possible medical conditions and common treatments. It aims to empower healthcare professionals by connecting them to individuals who need care the most, rather than those who can simply be diagnosed with AI.
Insitro: A San Fransisco-based company which uses advanced machine leanring with computational genomics to decrease the time and costs required for drug discovery.
Challenges of Implementing AI in Healthcare
Accuracy: AI systems need to be accurate, reliable, and valid, meaning that they can produce correct, consistent, and meaningful results that can be trusted and verified. However, AI systems may suffer from errors, biases, and uncertainties that can affect their performance and outcomes. For example, AI systems may have errors in their data, algorithms, or models that can lead to incorrect or misleading results, such as false positives or negatives, misdiagnosis, or adverse effects. AI systems may also have biases in their data, algorithms, or models that can lead to unfair or discriminatory results, such as favoring or excluding certain groups of patients or healthcare providers. AI systems may also have uncertainties in their data, algorithms, or models that can lead to ambiguous or uncertain results, such as conflicting or contradictory recommendations.
Privacy: Additionally, these machines need to protect the privacy, security, and consent of patients and healthcare data, meaning that they can respect and safeguard the personal and sensitive information of the patients and healthcare providers, and obtain their permission and agreement for using their data and providing them with services. However, AI systems may pose risks to the privacy, security, and consent of patients and healthcare data, such as exposing, leaking, or misusing their data, or violating their preferences, rights, or expectations. For example, AI systems may expose, leak, or misuse the data of patients or healthcare providers by hacking, stealing, or sharing their data without their knowledge or consent, or by using their data for purposes other than those intended or agreed upon, such as marketing, profiling, or surveillance. AI systems may also violate the preferences, rights, or expectations of patients or healthcare providers by providing them with services without their knowledge or consent, or by providing them with services that are not in their best interest or that are against their values or beliefs, such as unwanted, inappropriate, or harmful interventions.
Ethics: AI systems need to address the ethical, legal, and social implications of AI in healthcare, meaning that they can consider and comply with the moral, normative, and regulatory frameworks and standards that govern the healthcare domain, and that they can anticipate and mitigate the potential impacts and consequences of their use and deployment. However, AI systems may raise ethical, legal, and social issues and challenges that can affect the values, rights, and responsibilities of the stakeholders involved in healthcare, such as patients, healthcare providers, researchers, developers, regulators, and society at large. For example, AI systems may raise ethical issues and challenges such as respect for human dignity, autonomy, beneficence, non-maleficence, justice, and accountability. AI systems may also raise legal issues and challenges such as the liability, accountability, and responsibility of the actors involved in the development, deployment, and use of AI systems, as well as the protection and regulation of the intellectual property, data, and privacy rights of the stakeholders. AI systems may also raise social issues and challenges such as the impact on the social and cultural norms, values, and expectations of the healthcare domain, as well as the potential for social good or harm, inclusion or exclusion, empowerment or disempowerment, and trust or distrust of the AI systems.
How I Applied Image Recognition to Medical Imaging Data
My goal is to create impactful projects that can address real-world problems. I worked to achieve this goal through my Parkinson’s Disease Detector program, which uses image recognition to classify MRI scans of patients. I faced many challenges in obtaining, processing, and analyzing the data, as well as developing and evaluating the program.
First, I had to search for a suitable dataset of MRI scans for Parkinson’s disease, as there was no public one available. I contacted the University of Southern California, which had the largest repository of such scans, and applied for access to their Parkinson’s Progression Markers Initiative (PPMI) resource. I worked with them to be able to begin my project and after getting the approval, I downloaded the scans, which were in DICOM file format. I used MicroDicom, a free tool, to view and convert the files to JPEG format for easier access.
Next, I had to build the image recognition algorithm. I had little prior experience in this field, but I was eager to learn and implement my idea. I searched for tutorials on how to build image recognition models, but I found that most of them were outdated and incompatible with my project. I decided to learn the basics of image recognition from scratch. I also had to choose a machine learning framework to use for my project. I had two options: PyTorch, which was a simple and beginner-friendly application, and TensorFlow, which was a popular and advanced open source software library. I chose TensorFlow, as I wanted to challenge myself and achieve high standards in my work.
Meet ParkinSight
After a few weeks of working on this project, I finally built a functional program that could identify patterns within images and recognize them. I tested the program with the MRI scan images that I had obtained. However, I noticed that the accuracy of the model was only around 60%, which meant that the program could be making random guesses. I researched the possible reasons for the low accuracy and realized that my model had only 60 images to train and test with. I could not find any more images to use, so I had to work with what I had. I thought of augmenting the images by rotating them, changing their brightness levels, and making other adjustments to increase the number and variety of the images and data. This improved the performance of my model and enabled it to predict with over 95% accuracy whether a patient had Parkinson’s disease or not. If you would like to explore my code, please use this link (by choosing to utilize this tool, you consent to your data being processed for the diagnosis).
Watch the following tutorial if you would like to test out my program yourself:
The future of Healthcare with AI
Modern tech aims to eliminate manual patient reports for doctors, and aid physicians in diagnosis. Here are some of the future technologies we can expect to become a staple in the medical industry soon.
→ Patient Care Optimization: As AI Advances, more and more clinicians will have to learn to use AI to efficiently analyze patient histories and determine the optimize care strategies for each individuals. AI can help reveal hidden data trends that are difficult to predict, and this technology will enable clinicians to be able to work faster.
→ Resource Management: AI can be used to evaluate patient needs and manage hospital resources effectively. Providers are usually well-trained on the former but rarely the latter, which makes it challenging, especially given the strain on hospital capacity that’s so common these days.
Research shows that when patients who had a clinical need for admission to the ICU are instead admitted to another part of the hospital (e.g., a general ward), this results in longer hospital stays and higher readmission rates. This is not uncommon and thus it would be especially wise to hand such as a task to a computer, which can be much more proficient. New decision-support algorithms can be designed to incorporate tradeoffs of different options in these cases, can weigh the costs and benefits of the different choices, and provide appropriate recommendations.
→ Staffing Optimization: Attendance analytics can help to predict nurse absenteeism and optimize team composition. Algorithms can thus enhance consistency and predictability of work schedules, which will improve performance and reduce turnover rates.
→ Scheduling Enhancement: Machine learning predicts procedure durations and optimizes resource scheduling. These algorithms can prevent post-anesthesia care unit congestion and minimize delays. c
However, all of these technologies will require certain precursors for implementation. These include:
- Data Collection: Capture operational characteristics alongside clinical factors to make sure it functions consistently.
- Scalability and Interoperability: Design systems for scalability and interoperability.
- Human-Algorithm Interaction: Develop fair, explainable algorithms and incorporate human judgement where necessary.
Conclusion
An aging population, chronic conditions, and medical advances necessitate efficient hospital operations. AI has the potential to transform healthcare in many ways, from diagnosis and treatment to research and administration. These still-unrealized gains can improve healthcare outcomes, reduce healthcare costs, and empower healthcare professionals and patients.
However, AI also requires careful evaluation, regulation, and governance to ensure its safe, ethical, and responsible use in healthcare. AI in healthcare is not a replacement for human intelligence, but a complement and a tool that can augment and enhance human capabilities and performance. Through my Parkinson’s Disease Detector, I also hope to make an impact in the field of healthcare using AI.
Hey, I’m Shehraan, a 17-year-old driven to impact the world using emerging technologies. If you enjoyed my article or learned something new, feel free to follow me on Medium to keep up with my progress in Artificial Intelligence exploration and an insight into everything I’ve been up to. You can also connect with me on LinkedIn! Thank you so much for reading.
