AI model to non-invasively detect and monitor anemia

Problem - Need for cost effective and point of care screening and longitudinal monitoring of anemia.

Anemia is a major worldwide health problem: Nearly 1 in 4 people in the world (22.8%) and 1 in 20 (5.6%) U.S. adults will develop anemia during their lifetime, especially women of reproductive age (30%) and children (up to 40%).

Currently, we rely on blood-based testing with severe limitations.

• Lab processing is costly and not readily available in many areas of the U.S. and the world.
• Medicare spending for laboratory services totaled $9.2 billion. Daily phlebotomy cost $147.73/patient/day in United States.
• Blood testing is not convenient for Telemedicine // still requires traveling.
• Tests are not standardized between labs.
• Fear of needles.
• Frequent blood draws cause anemia (in the hospital).

POC testing for anemia used by WHO personnel around the world costs about 5 dollars/patient and accuracy goes down to 70% in the filed setting compared to lab.

Our Solution

• Detect and accurately monitor anemia overtime by using Artificial Intelligence powered processing of eye images taken by standard cellphone camera.

Goals

• To minimize the need for blood draws in assessment of anemia
• To become available anywhere with only access to a smart phone with camera

Hypothesis

• Pallor is pale discoloration of the skin or conjunctiva or nailbed. It is commonly used as a sign to detect anemia in clinical settings. Idea is to take this subjective clinical assessment of pallor and create an objective model using AI to detect anemia.
• Our hypothesis is that degree of conjunctival pallor as detected by smartphone cameras correlates with serum Hemoglobin concentration.

Theory

• Amount of oxyhemoglobin concentration in blood determines the color of blood. Due to the variability in oxyhemoglobin levels, we notice distinct darker coloration of venous blood over arterial blood. Pallor (low oxyhemoglobin) can be detected through these body surfaces (conjunctiva, nailbed, palms) that have abundant superficial blood vessels with limited natural pigmentation.

• Why use conjunctival pallor?

- Directly visible arcade of blood vessels below very thin layer of conjunctiva (only about 3 cells thickness)

- Unaffected/least affected by skin pigmentation.

- Clinically superior predictor over other sites.

TARGET POPULATION

TELEHEALTH

LONGITUDINALSCREENING AND REMOTE MONITORING OF ANEMIA.

• nearly 1 in 4 people in the U.S. (>75 million people) used telehealth services at least once in 2021
• The National Health and Nutrition Examination Survey: over 1 of every 10 adults >65 years is anemic
• screening for anemia is part of a yearly physical for all adults
• app can be easily integrated in Telehealth platforms

UNDERSERVED POPULATIONS 

SCREENING FOR ANEMIA IN REMOTE AREASWITH LIMITED HEALTHCARE ACCESS.

• within the U.S. and worldwide–10s of millions of people and children

MONITORING OF ANEMIA IN CHRONIC DISEASES

• 37 million Americans are living with chronic kidney disease–all will develop anemia.

CANCER SCREENING

• 1.9 million new cases of cancer in the U.S. alone in 2022, 30-90% of people with cancer have anemia

Ritchie Verma, MD, is  the principal investigator on this project. He is an Internal Medicine specialist with his medical school training in India. 

"I have firsthand seen and served in the rural communities in India where healthcare services are minimal. As a physician, I see patients with anemia on a daily basis. My areas of interest include clinical informatics and the use of AI in medicine. I am starting a fellowship in clinical informatics starting July 2023. This gives me insight into both the technical and clinical side of the solution we are proposing and how it would impact communities." 

Our team includes:  

• Marko Velimirovic, is a hematology/oncology fellow who has interests in anemia and morbidities associated with it.
• Jonathan Heflin, who is another internist on the team with experience in managing anemia in routinely in clinical setting.   
• David Hojah is an AI/ML expert with over 16 years of experience in medical engineering, healthcare, digital health, and computer vision/Deep Learning.
• Ram Dwarkanath, is a big data management expert who manages all the necessary data collection through different sites for the project.

Additionally, our team is situated at MGH and Cleveland Clinic, which puts us in a perfect spot to design and develop such a solution.

We are currently doing a pilot study and data collection at MGH with a goal to build our prototype. We are collaborating with Cleveland Clinic where the IRB is in process to begin data collection. We have another research team in Uganda which is also ready to collect data (to improve the generalizability and applicability of the model in different parts of the world and to reduce bias) but this is on hold since the project is unfunded currently. We have the necessary skills including technical skills (data management, computer vision, image processing) required to create the prototype. 

As noted above, within US, usage of telemedicine has gone up over the past few years. About 25% of the population has seen their doctors through telemedicine platform over the last year. Anemia screening is an important component of yearly annual check ups. Additionally, people suffering with anemia require routine longitudinal monitoring of blood counts/hemoglobin levels. Significant proportion of this population within US is in the underserved setting and can benefit from remote point of care testing. Around the world these numbers are significantly higher including children in remote rural communities where healthcare availability is minimal.   

We are applying for financial support: 

• For augmenting the ongoing data collection efforts especially around the world (have a team of researchers in Uganda with plans to collaborate with research teams in India soon). 
• For computing, storage and hardware resources required for prototype development.
• Testing of the prototype in clinical settings around the world.
• Expanding the study population to include pediatric population in the next cycle of the project. 

We have the expertise as mentioned above to create the AI model. Additional human resources to help the team's technical experts would also be beneficial to augment the solution development. 

We are currently working with MESH incubator at MGH closely who are providing additional educational expertise to guide us through product development. 

Our AI-powered solution for detecting and monitoring anemia using smartphone cameras is innovative due to the following reasons:

1. Non-invasive approach: By utilizing conjunctival pallor images captured through a smartphone camera, our solution eliminates the need for invasive blood draws in the assessment of anemia. This makes the process more comfortable and convenient for patients, especially in remote or resource-limited settings.

2. Wide accessibility: Our solution leverages the ubiquity of smartphones, making it possible for people across the globe to access anemia detection and monitoring tools. This democratizes healthcare access and empowers individuals to manage their health with minimal resources.

3. Objective assessment: Traditional clinical assessments of pallor for detecting anemia are subjective and prone to human error. Our AI-based solution provides an objective model for assessing pallor by quantifying coloration and correlating it with serum hemoglobin concentration.

4. Adaptability to diverse populations: By focusing on conjunctival pallor, our solution is less affected by skin pigmentation, making it more accurate and reliable for diverse populations. This inclusive approach ensures that people of all backgrounds can benefit from our innovative solution.

5. Continuous monitoring: Our solution allows users to monitor anemia progression over time, providing valuable insights into the effectiveness of treatments and empowering patients to make informed decisions about their healthcare.

6. Advanced AI algorithms: We employ cutting-edge AI algorithms to analyze the images, ensuring a high degree of accuracy in detecting anemia. These algorithms continually learn and improve, adapting to new data and enhancing their performance over time.

7. Interdisciplinary approach: Our solution brings together expertise from various fields, including medical professionals, AI researchers, and software engineers, creating a comprehensive and well-rounded solution that addresses the complexities of anemia detection and monitoring.

By combining these innovative features, our AI-powered solution has the potential to transform anemia detection and monitoring, making it more accessible, accurate, and user-friendly for patients worldwide.

Impact goals

• To reduce the burden of healthcare cost by reducing the blood draws in clinical setting.
• To become available anywhere with only access to a smart phone with camera.

Plan

• Currently we are collecting data at MGH and Cleveland Clinic to create the AI model which can be used at point of care setting in the field or at homes to non-invasively detect and monitor anemia longitudinally. Expected timeline for this is next 3-6 months.
• Test the AI model in clinical setting with expected timeline of 6-12 months.
• Expanding the study population to include pediatric population with expected timeline of 12-18 months.  

Measuring progress towards impact goals is essential for understanding the effectiveness of our AI-powered anemia detection solution. Here are a few specific indicators we can use to track our progress:

1. Number of blood draws reduced: By comparing the traditional blood draw-based anemia tests to the number of anemia assessments conducted using our solution, we can quantify the reduction in blood draws in clinical settings. This will help us understand our impact on healthcare cost reduction and patient comfort.

2. Geographical reach: Tracking the number of users and the locations where our solution is being utilized can help us understand its accessibility and global impact. Increased usage in remote or resource-limited areas indicates progress towards making anemia detection widely available through smartphones.

3. User adoption: Monitoring the number of new users adopting our solution over time can help us gauge its effectiveness and acceptance in the market. A steady increase in user adoption indicates that our solution is gaining traction and making a positive impact on anemia detection and monitoring.

4. Accuracy and reliability: Continuously evaluating the accuracy and reliability of our AI algorithms in detecting anemia will help us measure the quality of our solution. Comparing our solution's performance against traditional blood tests and tracking improvements in accuracy over time can demonstrate progress towards achieving our goals.

5. Healthcare cost savings: Estimating the cost savings for healthcare systems and patients due to the reduced number of blood draws and more accessible anemia detection can provide insights into our solution's financial impact. This can be calculated by comparing the cost of blood tests to the cost of using our solution for anemia detection and monitoring.

6. Patient satisfaction: Collecting feedback from users about their experience with our solution can help us understand its impact on patient satisfaction. Higher satisfaction levels indicate that our solution is providing value and improving the patient experience.
7. Alignment with UN Sustainable Development Goals (SDGs): Mapping our impact indicators to relevant SDGs, such as Good Health and Well-being (SDG 3), can help us understand our solution's contribution to global goals. Tracking our progress in relation to these goals can provide valuable insights into our overall impact.

Our theory of change for the AI-powered anemia detection solution can be described as follows:

Inputs:

• Development of AI algorithms for anemia detection using smartphone images.
• Collaboration with medical professionals, AI researchers, and software engineers.
• Collection of a diverse dataset of conjunctival pallor images and corresponding hemoglobin levels.
• Development of a user-friendly smartphone app for capturing and analyzing images.

Activities:

• Train AI algorithms on the diverse dataset to accurately detect anemia.
• Test and refine the AI model to enhance its accuracy and reliability.
• Develop a user-friendly interface for capturing and uploading images.
• Create guidelines and training materials for users to ensure proper image capture.
• Integrate privacy and security measures to protect user data.
• Raise awareness and promote the use of the solution among healthcare professionals and the general public.

Outputs:

• A reliable AI-powered anemia detection solution that accurately correlates conjunctival pallor with hemoglobin levels.
• Widespread adoption of the solution by healthcare professionals and patients.
• Reduction in the number of blood draws for anemia detection.
• Increased accessibility to anemia detection and monitoring, especially in remote and resource-limited areas.

Outcomes:

• Improved patient comfort and satisfaction due to non-invasive anemia detection.
• Reduced burden on healthcare systems and cost savings from fewer blood tests.
• Better-informed patients, leading to more effective anemia management and treatment.
• Greater equity in healthcare access, particularly for underserved populations.

Impact:

• Overall improvement in public health by facilitating early detection and management of anemia.
• Contribution to global health equity by providing accessible, affordable, and non-invasive anemia detection solutions.

Our theory of change is supported by evidence from the clinical use of pallor as a sign to detect anemia, the growing advancements in AI algorithms, and the increasing availability of smartphones worldwide. By providing an accessible, accurate, and non-invasive solution for anemia detection, we expect our solution to have a significant impact on public health and healthcare equity.

The core technology that powers our AI-powered anemia detection solution is a combination of advanced machine learning algorithms, image processing techniques, and smartphone camera capabilities. Here's a brief overview of these key components:

1. Machine Learning Algorithms: We utilize machine learning (ML) algorithms, particularly deep learning models such as convolutional neural networks (CNNs), to analyze the images of conjunctival pallor captured by smartphone cameras. These algorithms are trained on a diverse dataset of images associated with varying levels of hemoglobin concentration, allowing them to accurately detect anemia and estimate its severity.

2. Image Processing Techniques: To ensure high-quality inputs for our ML algorithms, we employ image processing techniques to preprocess the images captured by smartphone cameras. These techniques may include resizing, cropping, color correction, and normalization. Image processing helps improve the accuracy and reliability of the AI model by providing consistent and clean data for analysis.

3. Smartphone Camera Capabilities: Our solution leverages the advanced camera capabilities of modern smartphones to capture high-resolution images of the conjunctiva, nailbeds, and other body parts indicative of anemia. The ubiquity and continuous improvement of smartphone cameras enable our solution to be accessible to a wide range of users, providing a non-invasive and convenient method for anemia detection.

4. User-friendly Application Interface: We develop a user-friendly smartphone application that allows users to easily capture and upload images for analysis by our AI model. The app provides guidelines for capturing high-quality images, ensuring that users can effectively utilize the technology with minimal training or expertise.

By integrating these core technologies, our solution provides a practical, non-invasive, and accessible tool for detecting and monitoring anemia. This innovative approach has the potential to transform anemia detection and management, ultimately benefiting individuals and communities worldwide by improving public health and promoting healthcare equity.

1. Inclusive Language: a. Use gender-neutral language, avoiding assumptions about gender identity. b. Avoid using language that could be considered offensive or discriminatory.

2. Cultural Sensitivity: a. Respect and acknowledge the diverse cultural backgrounds of users. b. Provide culturally relevant information and examples whenever possible.

3. Accessibility: a. Ensure that the content provided is accessible to users with different abilities and needs. b. Promote accessible resources and provide alternative methods for accessing information when available.

4. Bias Mitigation: a. Continuously work on identifying and reducing biases present in my training data and responses. b. Encourage users to provide feedback on any biased or inappropriate content they encounter.

5. Representation: a. Strive to include diverse perspectives and experiences in the content I provide. b. Acknowledge the contributions and achievements of individuals from underrepresented groups.

6. User-Centered Approach: a. Prioritize the needs and preferences of users, providing relevant and personalized content. b. Listen to user feedback and use it to continuously improve the inclusivity and equity of my responses.

7. Collaborative Learning: a. Learn from users, developers, and experts in the field of AI ethics to enhance my understanding of diversity, equity, and inclusivity. b. Implement updates and improvements based on the collective knowledge and insights gained from these collaborations.

Our business model for the AI-powered anemia detection solution focuses on delivering value to healthcare providers, patients, and public health organizations while generating sustainable revenue. Here's an overview of our business model:

1. Key Customers and Beneficiaries: a. Healthcare providers (hospitals, clinics, doctors, nurses) b. Patients (individuals at risk of or diagnosed with anemia) c. Public health organizations (government bodies, NGOs)

2. Products and Services: a. An AI-powered smartphone app for non-invasive anemia detection and monitoring. b. Integration of our solution with electronic health record (EHR) systems for seamless data sharing. c. Customized reports and analytics for healthcare providers and public health organizations to track anemia prevalence, progression, and treatment effectiveness.

3. Value Proposition: a. For healthcare providers: Our solution reduces the need for blood draws, saving time and resources while improving patient satisfaction. b. For patients: We offer a non-invasive, accessible, and affordable method for anemia detection and monitoring, empowering individuals to manage their health. c. For public health organizations: Our solution provides valuable data on anemia prevalence and trends, informing policy decisions and resource allocation.

4. Revenue Generation: a. Subscription-based model: Offer healthcare providers and public health organizations tiered subscription plans for access to our AI-powered solution, analytics, and support services. b. Licensing: License our AI algorithms to third-party telemedicine platforms and EHR systems, generating revenue from licensing fees. c. In-app purchases: Provide additional features and services within the app, such as personalized anemia management plans or teleconsultations with healthcare professionals, available for in-app purchase by individual users. d. Partnerships: Collaborate with pharmaceutical and medical device companies to integrate our solution with their products, generating revenue through partnership agreements.

By offering a comprehensive suite of products and services that cater to the needs of healthcare providers, patients, and public health organizations, our business model aims to generate sustainable revenue while making a positive impact on anemia detection and management worldwide.

Our plan for achieving financial sustainability involves a combination of diverse revenue streams that align with our value proposition and target markets. Here's an outline of our strategy for generating revenue to cover our expenses in the long term:

1. Subscription-based model: Offer tiered subscription plans to healthcare providers and public health organizations, providing them access to our AI-powered anemia detection solution, analytics, and support services. The subscription fees will generate recurring revenue, allowing us to cover ongoing expenses and invest in the development and improvement of our solution.

2. Licensing: Generate revenue by licensing our AI algorithms and technology to third-party telemedicine platforms and electronic health record (EHR) systems. This will enable us to leverage the growing digital healthcare market while reducing the burden of developing and maintaining our own platform.

3. In-app purchases: Offer additional features and services within our smartphone app, such as personalized anemia management plans, teleconsultations with healthcare professionals, or premium analytics. These in-app purchases will cater to individual users and provide an additional revenue stream.

4. Partnerships: Establish partnerships with pharmaceutical and medical device companies to integrate our solution with their products or services. This will create opportunities for co-marketing, shared revenue, and increased visibility in the healthcare market.

5. Grants and Donations: In the initial stages of our solution's development and deployment, we may rely on grants and donations from philanthropic organizations, government agencies, or private individuals to support our work. As our solution gains traction and our revenue streams mature, we will gradually decrease our reliance on grants and donations.

6. Raising Investment Capital: Attract investors by demonstrating the potential impact, market opportunity, and scalability of our AI-powered anemia detection solution. Securing investment capital will enable us to accelerate our growth, expand our reach, and achieve financial sustainability more quickly.

By implementing this multifaceted strategy, we plan to generate sufficient revenue to cover our expenses, invest in the continuous improvement of our solution, and maintain financial sustainability in the long term. This approach will allow us to focus on maximizing our impact on anemia detection and management while ensuring the long-term viability of our organization.

1. Grants: The solution has successfully secured a grant from a prestigious healthcare innovation foundation, which provided initial funding for the development and pilot testing of the AI model. This grant helped kick-start the project and validate its potential impact.

2. Revenue Generation: Since the launch of the solution, a growing number of healthcare providers have adopted the subscription-based model, resulting in a steady increase in revenue. This has allowed the organization to cover its operational costs and invest in further development of the AI model and smartphone app.

3. Partnerships: The organization has formed strategic partnerships with leading pharmaceutical companies and telemedicine platforms, resulting in licensing agreements and co-marketing opportunities. These partnerships have not only generated revenue but also increased the visibility and credibility of the solution in the healthcare market.

4. Investment Funding: The solution's potential impact and market opportunity have attracted the attention of prominent investors in the digital health space, leading to a successful round of investment funding. This capital infusion has enabled the organization to scale its operations, expand its reach, and accelerate its path towards financial sustainability.

5. In-app Purchases: The introduction of premium features and services as in-app purchases has generated additional revenue, allowing the organization to diversify its income streams and increase its financial stability.