Table of Contents
Computer Vision and Deep Learning for Healthcare
Today, almost half of the world’s population does not have access to proper healthcare, with many people driven into poverty because of high health expenses. It is estimated that over $140 billion is required annually to meet the health-related sustainable development goal objectives. Further, significant health technology, digital technology, and artificial intelligence (AI) investments are needed to bridge the health service gap in emerging markets.
Many health-related startups and tech innovators have started integrating AI with their products and solutions, showing promise of improved diagnoses, reduced costs, and proper access to remote health services. COVID-19 has also accelerated the pace of transition to digital health applications, including those that integrate AI.
Health startups and tech companies aiming to integrate AI technologies account for a large proportion of AI-specific investments, accounting for up to $2 billion in 2018 (Figure 1). These investments range from digital diagnosis to clinician decision support to precision medicine. Accenture estimates that the health AI market in the United States is expected to grow at 40% annually.
This series is about CV and DL for Industrial and Big Business Applications. This blog will cover the benefits, applications, challenges, and tradeoffs of using deep learning in healthcare.
This lesson is the 4th of a 5-lesson course on CV and DL for Industrial and Big Business Applications 102.
- Computer Vision and Deep Learning for Oil and Gas
- Computer Vision and Deep Learning for Transportation
- Computer Vision and Deep Learning for Logistics
- Computer Vision and Deep Learning for Healthcare (this tutorial)
- Computer Vision and Deep Learning for Education
To learn about Computer Vision and Deep Learning for Healthcare, just keep reading.
Computer Vision and Deep Learning for Healthcare
Benefits
Unlocking Data for Health Research
The volume of healthcare-related data is increasing at an exponential rate. It has enormous potential to be mined and analyzed to facilitate better and more personalized care, reduce medical errors, and enable earlier disease diagnoses. However, having enough quality and well-structured data in emerging markets is challenging. Companies are tackling the weakness in data interoperability by cleaning and structuring data and overlaying analytics to make meaningful predictions to improve health.
In addition to structuring data for research, machine learning (ML) can match patients to clinical trials, speed up drug discovery, and identify effective life-science therapies when applied to big data. For example, the SOPHiA GENETICS AI technology computes one genomic profile every 4 minutes. It has analyzed hundreds of thousands of genomic profiles to facilitate matching patients to clinical trials across their research community.
Healthcare Efficiency
Software-as-a-service companies leverage data like patient history, consultation notes, diagnostic images, public information, and pharmaceutical prescriptions to automate multiple workflows like follow-up appointments. These services are impacting emerging markets by addressing the difficulties in finding appropriate providers or specialists, lack of transparency regarding quality of service, etc.
For example, China’s Ping An Good Doctor is one of the largest online platforms that use AI technologies to pre-triage patients to facilitate 24/7 online consultation services via its in-house medical team.
Reaching Underserved Communities
With a low supply of healthcare services, emerging markets have expressed a high willingness to engage with digital healthcare to get answers to health-related questions, diagnose a condition, and recommend treatments. This means a vast scope of digital health solutions can reduce the cost of reaching patient segmentations, particularly in rural areas.
The Indian platform DocsApp has developed a clinical AI platform called CLARA that connects patients to specialists and facilitates remote diagnoses and treatment. In addition, potential patients enter personal details and health concerns into the app.
Machine learning uses public data sources and customer information to generate a probable diagnosis and recommend a specialist. The patient then has the option to review the user ratings and credentials of the specialist and pay a fee online if they wish to proceed with the consultation via chat or call. The option to receive advice via SMS is a game-changer in lower-income and rural areas where the 3G-4G connectivity and smart-device penetration required for video consultations are missing or inadequate.
Other important roles of AI in healthcare are shown in Figure 2.
Applications
Medical Research in Genetics and Genomics
AI can benefit the medical research community by analyzing and identifying patterns in large and complex datasets to aid drug discovery, matching patients to clinical trials, and identifying effective life-science therapies. AI can also assist researchers in identifying genetic mutations that cause a particular disease and help predict the effects of treatments. For example, the Institute of Cancer Research cancer database combines genetic and clinical data from patients with information from scientific research. It uses AI to make predictions about new targets for cancer drugs.
Further applications in genomic medicine (e.g., oncology diagnosis and management) have facilitated the development of “precision medicine,” a data-driven approach to treatment that accounts for variability in genes, environment, and lifestyle factors to personalize medical care (Figure 3). AI can identify patterns and provide insights into how human physiology works and reacts to different chemicals, viruses, and the environment. Machine learning algorithms can also recognize patterns in DNA sequences and predict a patient’s probability of developing an illness.
These algorithms can design potential drug therapies, identify genetic causes of disease, and help understand the mechanisms underlying gene expression. They are also being used to discover biomarkers for schizophrenia, bipolar disorder, and depression to improve diagnosis or generate hypotheses for future research. AI can also perform data extraction, search systematic reviews, and assess health technology.
AI-enabled liquid biopsies (Figure 4) also enable physicians to better predict patient outcomes by determining whether their current therapy is optimal or an alternative treatment may benefit a patient. Liquid biopsies analyze DNA from a blood sample and remove the need for invasive methods. One such innovation intends to predict a relapse 7 months earlier than the current standard of care in patients diagnosed with cancer.
AI may also improve gene editing accuracy (a method of altering DNA at the cellular or organism level). In the future, using large datasets and machine learning may predict optimal locations to edit DNA to alleviate suboptimal gene editing outcomes, enabling researchers to focus efforts on genes that are less likely to be at risk to patients. Editing the egg, sperm, or embryo genome can also prevent the genetic risk of developing a disease, thus lowering the risk of infection even before birth.
Clinical Care
Medical Imaging and Radiology
Machine learning and deep learning are being used in radiology (Figure 5) to create tools that can improve the diagnosis and classification of cancer, breast cancer, and polyp detection in CT colonography. In addition, deep learning algorithms can automate the extraction and classification of images with speed and power, assisting in diagnosing stroke using neuroimaging with a CT and MRI.
Super-resolution-based AI algorithms can also improve the image quality of scans, which are often lower quality because of the time constraints in managing stroke patients. When combined with X-ray and MRI, AI can automatically delineate tumors and improve tuberculosis diagnosis. With positron emission tomography (PET), it can assist in the early diagnosis of Alzheimer’s disease.
AI can act as a decision support tool in endoscopy to predict the pathology of lesions and prevent unnecessary polypectomy of non-cancerous polyps. In addition, concurrent developments in medical imaging relate to AI tools intended to improve patient and clinician safety and patient experience. These include applications that enable CTs to be performed at ultra-low radiation doses, MRI exams that can be conducted in two-thirds of current time frames, and PETs that use radiotracer dose reductions of up to 99%.
Pathology
Digital pathology (Figure 6) has created large volumes of data to train AI algorithms that recognize patterns and help balance the world’s shortage of pathologists. To assist pathologists, AI can be leveraged to
- Automate complex and time-consuming tasks like object quantification, tissue classification based on morphology, and target identification
- Calculate personalized treatments using the available data
- Minimize the risk of misdiagnosis and wrong prescription of medicine
- Promote telepathology by allowing physicians in rural areas to consult specialized pathologists
- Identification of visually discernible markers by the human eye (e.g., molecular markers in tumors)
Dermatology
AI can support dermatologists in making clinical decisions for general skin and specific cancers (Figure 7). Most of these applications aim to diagnose and prevent skin disease onset. This will minimize unnecessary biopsies and help clinicians to assist in the detection of skin disease. These AI algorithms learn to distinguish malignant melanomas from benign lesions and, at the same time, identify new lesion characteristics helpful in identifying cancer, which otherwise will be difficult to determine using visual diagnosis. By early-stage detection, machine learning can analyze the changes and development of skin moles. Individuals can scan their bodies using an AI application to find suspicious marks. AI is also used to diagnose acne, psoriasis, seborrheic dermatitis, and nail fungus.
Neurology
Besides neuroimaging, AI has excellent applications in neuroscience to support large-scale hypothesis generation and provide insights into the brain’s interactions, structure, and mechanisms (Figure 8). For example, these algorithms can recognize early warning signs of stroke by distinguishing normal resting and stroke-related paralysis. They can also predict 3-month outcomes of patients suffering from ischemic stroke by examining the patterns between physiological parameters.
Researchers have shown the capability of AI to assess the severity of Parkinson’s disease (PD). AI is also used with patients suffering from spinal cord injury by providing prognostic evaluation and predicting outcomes. In addition, algorithms are developed that provide data on patient-specific motor defects to be used with a robot-assistive rehabilitation harness to assist people in learning to walk again. It has also been used with electrical stimulation systems in quadriplegic patients to restore some movement.
Mental Health
Mental health is important for the proper functioning of human beings. Because of its sensitive nature, managing mental health is more effective when the person receiving care interacts with the healthcare provider. Natural language processing (NLP) algorithms and machine learning can gather and adapt to new information that can help healthcare providers stimulate participant-clinician interactions. Conversational agents, chatbots, and virtual assistants can mimic human-like presence. They can help enhance the searchability of online support communities, diagnose a major depressive disorder, and deliver cognitive behavior therapy to people with depression and anxiety.
These virtual agents can also act as moderators of an online community for youth mental health when human moderators are not available. Such agents can assess the sentiment, emotion, and keywords of participant posts that were used to recommend appropriate steps and actions.
Diabetes
Type 1 and Type 2 diabetes are prevalent in the population today, and because of that, large amounts of data about blood sugar and trends are readily available. Deep neural networks and support vector machines are being explored in developing pre-diabetic screening tools. The tool uses data from the Korean national survey that defines nine variables like family history, waist circumference, and physical activity to predict diabetes (e.g., Diabetic Retinopathy, see Figure 9).
Moreover, AI is being explored in the artificial pancreas system to support computer programs that connect the continuous glucose monitor to an insulin pump. This ensures personalized insulin delivery as AI algorithms can learn from data in uncertain environments. For example, compilations with diabetes can trigger other physical and heart-related diseases. AI can also be used to predict these compilations while they are still treatable. These models are trained on data from existing surveys that assess the relationship between risk factors like hemoglobin A1C.
Eye Care
AI can replace existing vision programs by allowing for point-of-care diagnostics of patients. Deep learning is being used to differentiate healthy eyes from eyes with age-related macular degeneration. It can predict cardiovascular disease from retinal fundus images, automate grading age-related macular degeneration, screen for glaucoma, and diagnose cataracts.
Challenges
Figure 10 lists the risks and challenges associated with using AI in the healthcare industry. The following are the most important.
Regulatory Friction
Healthcare is a high-stake game and has good reasons for rigorous regulatory frameworks. Yet, in several countries, the governance and legal frameworks for virtual care providing diagnosis and treatment are untested or underdeveloped. COVID-19 led several countries to temporarily waive limits on telehealth, while others turned a blind eye, given the urgency and necessity for remote care options during the pandemic.
AI innovations that feed on patient data must also navigate several data regulations, from storage to security and interoperability. This creates a tradeoff in balancing high standards like patient consent, privacy, and protection with the need for large structured datasets for training new AI applications that make healthcare personalized, efficient, and preventive.
Non-Representative AI: Potential Bias and Misdiagnosis
As discussed in previous points, privacy restrictions within healthcare organizations limit the size of the structured dataset that can be used for training AI algorithms. Further, because of this limitation, the resulting data cannot be scaled across markets as a population with different ethnic origins may have other predispositions for disease.
Further, using machine learning in clinical diagnostic applications has several risks that researchers are still grappling with. Some of the dangers in applying machine learning to clinical medicine include the following:
- The distributional shift is caused by the data distribution difference on which the system is trained and the data used in operation. This shift can arise because of changing disease patterns over time. Also, AI might not work well for the data that are scarce or more difficult to collect, such as for rare medical conditions and underrepresented communities like Black, Asian, and minority ethnic populations.
- AI systems are often black-box decision-makers, where it can be difficult or impossible to determine the underlying logic that generates the outputs produced by AI. This creates problems in validating AI systems’ outputs and identifying errors or biases in the data.
- Insensitivity to impact is a system designed to make accurate decisions at the cost of either missed or overdiagnosis. This is a dilemma that human clinicians are trained to address with judgment.
- Harmful unintended consequences are caused by a system trained only on historical data or using irrelevant data points that miss important predictive factors, resulting in missed or inaccurate diagnoses or overdiagnosis.
Effect on Patients and Healthcare Professionals
AI systems can restrict the choices based on risk or what is in the user’s best interests, thus negatively impacting individual autonomy. If healthcare professionals cannot explain how AI systems make diagnoses, this can weaken patients’ trust in the system. Applications that aim to imitate humans/doctors will raise the possibility that the user will be unable to judge whether they are communicating with a natural person or with technology. This could be experienced as a form of deception or fraud.
On the other hand, healthcare professionals may feel that their autonomy and authority are threatened if AI challenges their expertise. In addition, the ethical obligations of healthcare professionals toward individual patients might be affected by the use of AI decision support systems. Further, with the introduction of AI, professionals likely have to pick up new skills and expertise. Another concern is that AI could make healthcare professionals complacent and less likely to check results and challenge errors.
Institutional Inertia
Commercialization of healthcare innovations will require regulatory and institutional ecosystems to facilitate collaboration with academia, venture capitalists, angel investors, and entrepreneurs. Other geographically concentrated investment flows will mean that countries with significant health gaps will have more hurdles in developing and diffusing appropriate solutions for healthcare options.
Small innovators must partner with large digital platforms to leverage existing technological innovations to foster the diffusion of healthcare products in emerging markets. Further, the development of financial institutions and governments is needed to play an essential role in distribution by encouraging best-practice governance arrangements and investment in the human capital required to embed AI within new and existing healthcare organizations.
What's next? We recommend PyImageSearch University.
84 total classes • 114+ hours of on-demand code walkthrough videos • Last updated: February 2024
★★★★★ 4.84 (128 Ratings) • 16,000+ Students Enrolled
I strongly believe that if you had the right teacher you could master computer vision and deep learning.
Do you think learning computer vision and deep learning has to be time-consuming, overwhelming, and complicated? Or has to involve complex mathematics and equations? Or requires a degree in computer science?
That’s not the case.
All you need to master computer vision and deep learning is for someone to explain things to you in simple, intuitive terms. And that’s exactly what I do. My mission is to change education and how complex Artificial Intelligence topics are taught.
If you're serious about learning computer vision, your next stop should be PyImageSearch University, the most comprehensive computer vision, deep learning, and OpenCV course online today. Here you’ll learn how to successfully and confidently apply computer vision to your work, research, and projects. Join me in computer vision mastery.
Inside PyImageSearch University you'll find:
- ✓ 86 courses on essential computer vision, deep learning, and OpenCV topics
- ✓ 86 Certificates of Completion
- ✓ 115+ hours of on-demand video
- ✓ Brand new courses released regularly, ensuring you can keep up with state-of-the-art techniques
- ✓ Pre-configured Jupyter Notebooks in Google Colab
- ✓ Run all code examples in your web browser — works on Windows, macOS, and Linux (no dev environment configuration required!)
- ✓ Access to centralized code repos for all 540+ tutorials on PyImageSearch
- ✓ Easy one-click downloads for code, datasets, pre-trained models, etc.
- ✓ Access on mobile, laptop, desktop, etc.
Summary
Significant investments are being done in health technology and artificial intelligence to bridge the health service gap in emerging markets in the following ways.
- Medical Research: AI can benefit the medical research community by analyzing and identifying patterns in large and complex datasets to aid drug discovery, matching patients to clinical trials, and identifying effective life-science therapies.
- Clinical Care: Ranging from Radiology, Medical Imaging, Neurology, Diabetes, and Mental Health, AI has shown promising results.
However, AI in transport comes with its challenges.
- Regulatory Friction: Healthcare is a high-stake game with good reasons for rigorous regulatory frameworks. Yet, in several countries, the governance and legal frameworks for virtual care providing diagnosis and treatment are untested or underdeveloped.
- Non-Representative AI: Limited data cannot be scaled across markets as a population with different ethnic origins may have other predispositions for disease. This causes a shift where AI might not work well for the data that are scarce or more difficult to collect, such as for rare medical conditions and underrepresented communities (e.g., Black, Asian, and minority ethnic populations).
- Effect on Patients and Healthcare Professionals: AI systems can restrict the choices based on risk or what is in the user’s best interests, thus negatively impacting individual autonomy. If healthcare professionals cannot explain how AI systems make diagnoses, this can weaken patients’ trust in the system.
- Institutional Inertia: Geographically concentrated investment flows will mean that countries with significant health gaps will have more hurdles in developing and diffusing appropriate solutions for healthcare options.
I hope this post helped you understand the benefits, applications, challenges, and tradeoffs of using deep learning in healthcare. Stay tuned for the last lesson, where we will discuss deep learning and computer vision applications for education.
Consultation Services
Do you need some help with your industrial application of AI? Learn more about our consulting services.
Citation Information
Mangla, P. “Computer Vision and Deep Learning for Healthcare,” PyImageSearch, P. Chugh, A. R. Gosthipaty, S. Huot, K. Kidriavsteva, and R. Raha, eds., 2023, https://pyimg.co/h52u4
@incollection{Mangla_2023_CVDLH, author = {Puneet Mangla}, title = {Computer Vision and Deep Learning for Healthcare}, booktitle = {PyImageSearch}, editor = {Puneet Chugh and Aritra Roy Gosthipaty and Susan Huot and Kseniia Kidriavsteva and Ritwik Raha}, year = {2023}, note = {https://pyimg.co/h52u4}, }
Unleash the potential of computer vision with Roboflow - Free!
- Step into the realm of the future by signing up or logging into your Roboflow account. Unlock a wealth of innovative dataset libraries and revolutionize your computer vision operations.
- Jumpstart your journey by choosing from our broad array of datasets, or benefit from PyimageSearch’s comprehensive library, crafted to cater to a wide range of requirements.
- Transfer your data to Roboflow in any of the 40+ compatible formats. Leverage cutting-edge model architectures for training, and deploy seamlessly across diverse platforms, including API, NVIDIA, browser, iOS, and beyond. Integrate our platform effortlessly with your applications or your favorite third-party tools.
- Equip yourself with the ability to train a potent computer vision model in a mere afternoon. With a few images, you can import data from any source via API, annotate images using our superior cloud-hosted tool, kickstart model training with a single click, and deploy the model via a hosted API endpoint. Tailor your process by opting for a code-centric approach, leveraging our intuitive, cloud-based UI, or combining both to fit your unique needs.
- Embark on your journey today with absolutely no credit card required. Step into the future with Roboflow.
Join the PyImageSearch Newsletter and Grab My FREE 17-page Resource Guide PDF
Enter your email address below to join the PyImageSearch Newsletter and download my FREE 17-page Resource Guide PDF on Computer Vision, OpenCV, and Deep Learning.
Comment section
Hey, Adrian Rosebrock here, author and creator of PyImageSearch. While I love hearing from readers, a couple years ago I made the tough decision to no longer offer 1:1 help over blog post comments.
At the time I was receiving 200+ emails per day and another 100+ blog post comments. I simply did not have the time to moderate and respond to them all, and the sheer volume of requests was taking a toll on me.
Instead, my goal is to do the most good for the computer vision, deep learning, and OpenCV community at large by focusing my time on authoring high-quality blog posts, tutorials, and books/courses.
If you need help learning computer vision and deep learning, I suggest you refer to my full catalog of books and courses — they have helped tens of thousands of developers, students, and researchers just like yourself learn Computer Vision, Deep Learning, and OpenCV.
Click here to browse my full catalog.