Liver fibrosis early detection at low cost

In July 2021, the World Health Organization reported over 90 million people in Africa suffering from chronic hepatitis. Each year, more than 200,000 Africans die from hepatitis-related complications, primarily from undetected or untreated cirrhosis. These figures continue to rise [10].

In Senegal, hepatitis B has become endemic. Indeed, eighty-five percent (85%) of the population has been exposed to the Hepatitis B virus. Ten to eleven percent (10-11%) are chronic carriers, of whom 20% to 30% will progress to cirrhosis. Consequently, diagnosing liver cirrhosis due to hepatitis B is a significant public health issue in Senegal, often recognized too late [11].

Invasive tests such as histology following a biopsy are used to identify hepatic cirrhosis. These involve taking liver samples from the patient, which are then analyzed by pathologists.

In recent years, new non-invasive methods such as the Fibrotest, a blood marker, and Fibroscan, which measures liver stiffness using waves, have been developed to diagnose hepatic fibrosis or cirrhosis, offering an alternative to traditional biopsy. However, in Senegal, access to these innovative technologies remains limited. They are mainly available in two private laboratories and specialized institutions like the UFR of Health Sciences at the University of Thiès and the Infectious Diseases Department at the National University Hospital Center of Fann. Furthermore:
- The financial cost of the Fibrotest remains relatively high and unaffordable for most of the population, unlike the Fibroscan.
- Blood samples must be sent abroad, significantly impacting the time it takes to receive test results and consequently delaying the detection of cirrhosis and the quality of care.

In the follow-up of patients with the Hepatitis B virus, an abdominal ultrasound is periodically prescribed to monitor the liver and watch for the onset of cirrhosis. This examination is relatively accessible to the population. However, the resolution of the ultrasound does not allow for the early detection of cirrhosis. Moreover, elastography has proven effective in diagnosing hepatic fibrosis. In this context, it is important to consider the viscosity to make an accurate estimation of elasticity since it can also provide other mechanical information about the liver [12]. Additionally, regarding fibrosis (according to the Metavir score), studies analyzing 40,405 radiological images from 15,853 patients have shown an accuracy of over 85% compared to histopathological examination [13].

Despite the continuous decline in the incidence of hepatitis B and the control of hepatitis C, hepatic cirrhosis remains a public health issue associated with various complications and high mortality. Indeed, another formidable challenge to consider in clinical practice due to the interplay of NCDs (non-communicable diseases) includes alcoholic and non-alcoholic hepatic steatosis and autoimmune and drug-induced liver pathology [14].

These challenges underscore the need for an effective solution for the early detection of cirrhosis, such as the development of innovative technologies in this field.

Utilizing state-of-the-art artificial intelligence (AI) and deep learning techniques, our team has developed a cutting-edge proof of concept, alongside a prototype, that showcases the innovative use of ultrasound imaging for early liver cirrhosis detection. Our advanced prototype processes ultrasound images of the liver, harnessing the power of convolutional neural networks and transfer learning, to determine the degree of fibrosis or detect the onset of hepatic cirrhosis.

The prototype, accessible both as a mobile and web application, can be found at [fibrosedetection-u7vsmwkbgwyyevx32vh3o2.streamlit.app](https://fibrosedetection-u7vsmwkbgwyyevx32vh3o2.streamlit.app/). Furthering our research, we have disseminated our findings in a scholarly paper available at [Springer Link](https://link.springer.com/chap).

Our ambition is to elevate our model's capabilities by integrating advanced deep learning methodologies such as transformers and object detection. Additionally, we aim to augment our dataset, enhancing the model's robustness and interpretability.

Our vision extends to deploying this application on mobile devices, ensuring it is functional in both urban and rural areas equipped with the necessary ultrasound technology for liver fibrosis screening.

Ultimately, we aspire to revolutionize the diagnostics process by synergizing mobile ultrasound technology with our AI application, thereby facilitating rapid and informed decision-making for healthcare professionals in the fight against liver fibrosis.

The innovation aims to enhance social health outcomes by increasing liver fibrosis awareness and early detection, reducing mortality, expanding screening in underserved areas, and empowering patients with self-management tools over the next three years. It is expected to impact approximately at least one million individuals per year, based on WHO data on hepatitis prevalence in Africa and Senegal. The anticipated benefits include improved health and quality of life, greater access to screening, and healthcare cost savings through early intervention and decreased need for medical treatment abroad.

We initiated this project by engaging with healthcare professionals to discuss how AI tools could enhance their work. From these discussions, we identified the early detection of liver fibrosis as a critical challenge within the African context.

Following this, we developed a research protocol and secured a grant that enabled us to collect data directly from patients, obtaining their consent in the process.

We aim to expand this approach by employing a mobile ultrasound machine, which will facilitate our outreach to communities and improve patient access to diagnostic services.

Liver cirrhosis is a disease with significant global prevalence. Often termed a "silent disease," it progresses asymptomatically to a stage of decompensated cirrhosis. Despite advances in medicine, the mortality rate from this disease continues to rise. Therefore, it is crucial to leverage data science and artificial intelligence to enhance the diagnosis and identification of liver fibrosis, thereby facilitating early patient management and improving survival chances.

In this context, we propose a diagnostic tool that utilizes an automated transfer learning model. This model processes ultrasound images of the liver and outputs the corresponding level of fibrosis or liver cirrhosis.

Our initial solution employs convolutional neural networks, specifically the ResNet50 model, which is pre-trained on ImageNet data. We have evaluated its performance using several metrics, achieving 84% accuracy and 80% precision. The area under the curve (AUC) for the receiver operating characteristic (ROC) curve is 0.85.

These results have been documented in the Springer proceedings. A second iteration of the tool is currently under testing, which incorporates Transformers, and we have deployed a prototype on the web using Streamlit .

Our team comprises Artificial Intelligence faculty and students, along with healthcare professionals from LeDantec Hospital in Senegal. We plan to expand this project to Thies Hospital, collect more data, and accurately detect all four stages of fibrosis.

We shared the prototype with health professionals for validation but there's no used with patients. We intend to go further this year by acquiring mobile ultrasound machine and start the validation process.

We have shared the prototype with health professionals for initial validation, but it has not yet been used with patients. This year, we plan to acquire a mobile ultrasound machine and begin the comprehensive patient validation process.

Our team, composed of academics and students based in Senegal, has produced scientific publications and is confident in the impact our application can have in Africa and other developing regions. Our solution aims to save lives cost-effectively. To achieve this, we must transition from our academic environment to the world of social entrepreneurship—a shift that presents unique challenges, particularly in Francophone countries. I am applying to Solve to connect with global partners, learn about best practices in social entrepreneurship, and successfully launch our product in the market.

The second challenge involves the nature of our application—an AI-driven health tool. We face several hurdles, including acceptance by healthcare professionals, real-world application use, and ongoing assessment of our solution's robustness. Engaging with Solve partners will be crucial to address these challenges and refine our approach.

Our  AI solution for early detection of liver fibrosis in Africa is innovative for several reasons:

1. Access to healthcare: In Africa and other regions in the world, the availability of specialized healthcare services and diagnostic tools, especially for critical organs like the liver, remains a challenge. Our innovative AI solution is specifically designed for early detection of liver fibrosis at low cost. This tool can be seamlessly integrated into primary care facilities or utilized through remote access, significantly enhancing the reach and effectiveness of medical diagnostics. By providing  reliable support in diagnosis, it empowers healthcare providers to make informed decisions promptly, thereby improving patient outcomes in communities where healthcare resources are scarce.

2. Cost-Effectiveness: Advanced medical diagnostics in liver fibrosis are fibrotest and fibroscan and they are expensive and not affordable for many. Our AI solution can provide a cost-effective alternative, reducing the need for expensive tests and specialist consultations based just on ultrasound images.

3. Early Detection and Intervention: Our AI solution can analyze medical images or other data with high accuracy, potentially identifying fibrosis at an early stage when treatment can be more effective, thereby improving patient outcomes.

4. Resource Optimization: Healthcare resources in Africa are often stretched thin. Our AI solution can assist in prioritizing cases, managing patient care, and optimizing the use of scarce medical resources.

5. Training and Empowerment: Our AI solution can serve as a decision support tool for healthcare providers, enhancing their ability to diagnose and treat liver diseases, which may also reduce the need for extensive training in diagnostic imaging.

6. Research and Data Collection: In Africa and other regions globally, the availability of specialized healthcare services and diagnostic tools, especially for critical organs like the liver, remains a challenge. Our innovative AI solution is specifically designed to transform the early detection of liver fibrosis. This tool can be seamlessly integrated into primary care facilities or utilized through remote access, significantly enhancing the reach and effectiveness of medical diagnostics. By providing swift and reliable support in diagnosis, it empowers healthcare providers to make informed decisions promptly, thereby improving patient outcomes in communities where healthcare resources are scarce.

7. Adaptability: Our AI solution can be trained on local datasets, which means they can be adapted to recognize patterns and presentations of liver fibrosis that are specific to local population.

8. Scalability: Our AI solution can be scaled across multiple sites and regions, providing widespread benefits without the need for a proportional increase in human or physical resources.

Overall, our AI solution for the early detection of liver fibrosis represents a leap forward in leveraging technology to address healthcare challenges in Africa and in the world, particularly in regions where traditional healthcare infrastructure is lacking or under-resourced.

In simple terms, our AI solution for early liver fibrosis detection can be considered as a smart assistant that helps doctors spot liver problems much earlier than usual. It can make a big difference in curative or preventive health policies in Africa:

1. Reach: Many places in Africa don't have enough specialists who can diagnose liver diseases early. Our AI tool can be used by healthcare workers in local clinics, meaning more people can get checked out without having to travel far.

2. Speed and mobility: Our AI Solution will be deployed as a mobile app, can quickly analyze liver ultrasound images and spot any signs of fibrosis. This means patients can get a diagnosis much faster, without waiting for a specialist's appointment which can sometimes take a long time or data sent abroad. It can be coupled with a mobile ultrasound engine and help for liver fibrosis detection in rural zone without the need of a specialist.

3. Affordability: Specialized tests for the liver can be expensive. Our AI solution is a more affordable option and depends only on Ultrasound medical liver images, making it possible for more people to get tested even if they can’t pay much.

4. Education: The AI tool can also help teach local healthcare workers about liver diseases. By showing them what early fibrosis looks like, they can learn to recognize the signs themselves.

5. Data Gathering: Our  AI solution can collect data about liver health from different parts of Africa. This information is useful for understanding how liver diseases affect different communities and can help in planning better healthcare strategies.

6. Improving or saving lives: By catching liver fibrosis early, treatment can start sooner, which can save lives and improve the quality of life for many people.

 7. Prevention: Our AI solution can help making prevention campaign about liver fibrosis at low cost.

When considering the impact goals for our AI solution for liver fibrosis early detection, we would focus on several key areas, and it's important to establish metrics to measure progress effectively:

1. Increased liver fibrosis early diagnosis rates:    - Goal: To detect liver fibrosis at an earlier stage when interventions can be more effective.    - Metric: Number and percentage increase in early-stage diagnoses compared to baseline data.

2. Improved access to liver fibrosis diagnostics:    - Goal: To make liver fibrosis diagnostics accessible in remote and under-resourced areas.    - Metric: The number of healthcare facilities using the AI tool and the geographical spread.

3. Reduction in liver disease mortality:    - Goal: To decrease the mortality rate associated with liver diseases, especially in areas where medical intervention is scarce.    - Metric: Mortality rates from liver diseases before and after the implementation of the AI tool.

4. Cost-Effectiveness:    - Goal: To reduce the overall cost of liver disease diagnostics and treatment.    - Metric: Comparative cost analysis of liver fibrosis detection before and after the AI solution implementation.

5. Enhanced healthcare worker training:    - Goal: To provide healthcare workers with advanced tools and knowledge to manage liver diseases effectively.    - Metric: Number of healthcare workers trained and the subsequent impact on their diagnostic capabilities, assessed through skill evaluations.

6. Quality of life improvements:    - Goal: To improve the quality of life for patients with liver disease through earlier and more accurate diagnosis.    - Metric: Patient quality of life surveys or indices before and after diagnosis with our  AI solution.

7. Healthcare system strengthening:    - Goal: To bolster the overall healthcare system's response to liver diseases.    - Metric: Reduction in time from symptom onset to diagnosis and treatment, and the system’s capacity to handle liver disease cases.

8. Research and Development:    - Goal: To collect data that can drive further research into liver diseases specific to the African context.    - Metric: Number of research studies or papers published using data generated by the AI tool.

9. Patient engagement and education:    - Goal: To increase awareness and understanding of liver health among the population.    - Metric: The level of patient engagement in preventative health measures pre-intervention and post-intervention.

Progress towards these goals is typically tracked through regular reporting, data analysis, patient outcomes, and feedback from both healthcare providers and patients. Establishing a robust data collection and analysis framework is essential for ongoing monitoring and adjusting strategies to ensure that our  AI solution has a great impact.

Our technology harnesses the power of artificial intelligence within the realm of computer vision to enhance the precision of medical diagnostics. Our approach involves utilizing a robust dataset of ultrasound liver images complemented by clinical data to develop a sophisticated AI application capable of discerning various stages of liver fibrosis. 

Initially, our journey began with the implementation of Convolutional Neural Networks (CNNs), a class of deep neural networks well-suited for analyzing visual imagery. As we progress, our focus has shifted towards building an advanced model that employs transformers, an architecture that has revolutionized the field of natural language processing and is now proving to be equally transformative in image recognition tasks.

Our objective is to refine this model to accurately identify the four distinct stages of fibrosis, enhancing early detection and treatment. However, the acquisition of additional data is critical to achieve this level of granularity and reliability in our AI model. Hopefully we got this month a small grant from Melinda and Bill Gates foundation that will help us collect and annotate data.

To ensure the clinical validity and applicability of our technology, we have assembled a dedicated team of gastroenterology experts. These specialists are integral to our validation process, providing essential insights that steer the development of our AI application towards a tool that is both medically sound and practically useful in a clinical setting.

1. Full time staff
   1. Professor Mamadou Bousso CEO EDULYTICS
   2. Professor Madoky Diop Gastroentorologue
   3. Professor Ousmane SALL
2. Contractors
   1. Yakhoub Mass AI student
   2. Oumar Kane AI student
   3. Coumba Gueye ( Health student)
   4. Mouhamed Mbodj ( Health student)

We have been diligently working on this project since 2021. Over the course of one year, we gathered data and developed an initial prototype leveraging a convolutional neural network. Currently, we are in the second phase of the project, where we are enhancing the AI model by incorporating transformers. Additionally, we are expanding our data collection efforts to include new hospitals, rural areas, and various stages of liver fibrosis.

We have assembled a distinguished team that epitomizes academic excellence and professional expertise. Each member possesses a profound understanding and practical experience in the realms of AI and liver fibrosis, ensuring our collective proficiency in advancing our project.

The business model for our AI app dedicated to early detection of liver fibrosis needs to balance providing significant health benefits with generating sustainable revenue. Here's a breakdown of how such a business model might be structured:

Value Proposition The AI app provides critical value through early detection of liver fibrosis, which can significantly enhance treatment outcomes and reduce healthcare costs by preventing progression to more severe liver diseases like cirrhosis

Key Customers and Beneficiaries - Healthcare Providers: Hospitals, clinics, and medical professionals who need accurate, quick, and cost-effective diagnostic tools. - Patients: Individuals at risk of liver diseases who benefit from early diagnosis and management of their condition. - Healthcare Systems: National and regional healthcare systems interested in reducing the burden of liver diseases and associated costs. - Insurance Companies: Entities that can reduce payouts for advanced liver disease treatments by promoting early detection and management. - Research Institutions: Organizations looking to gather data on liver fibrosis to improve treatment protocols and outcomes.

Products or Services Offered

- Diagnostic Services: The app analyzes ultrasound images to detect liver fibrosis stages, offering a non-invasive and immediate diagnostic tool. - Data Analytics Services: Leveraging collected data to provide insights into liver disease patterns, treatment outcomes, and risk factors. - Subscription Models: For healthcare providers to get regular updates, enhanced features, and ongoing customer support. - Training and Support: Educating healthcare professionals on using the AI tool effectively.

- Ultrasound medical engine with embedded AI application for a near future

Revenue Streams - Subscription Fees: Charging healthcare facilities a monthly or annual fee to use the app. - Per-Use or Licensing Fees: Charging per diagnostic session or through licensing agreements with medical device companies or healthcare providers. - Data Monetization: Anonymizing and aggregating data to sell to research institutions or pharmaceutical companies for research purposes. - Partnerships and Grants: Collaborating with government bodies or non-profits for funding aimed at improving healthcare outcomes in underserved areas.

Delivery Channels - Direct Sales to Hospitals and Clinics: Integrating our AI app into existing diagnostic equipment or as a standalone application. - Online Platforms: Offering the app through cloud-based platforms for ease of access and updates. - Partnerships with Medical Device Manufacturers: Embedding the software in ultrasound and other imaging devices.

Customer Relationships - Dedicated Support: Providing training and technical support to ensure smooth operation and integration. - Community Building: Creating a network of users who can share insights, data, and best practices. - Feedback Mechanisms: Regularly gathering user feedback to improve and adapt the product.

Key Activities - Continuous AI Training and Development: To enhance accuracy and adapt to new data. - Marketing and Outreach: To promote widespread adoption and awareness of the product. - Compliance and Certification: Ensuring the app meets all medical and legal requirements for medical software.

By focusing on these areas, our AI app can provide significant value to both healthcare providers and patients, driving adoption and generating revenue while improving healthcare outcomes in the realm of liver fibrosis detection.

This project initially began as an unfunded academic research endeavor. However, as we advanced and demonstrated significant progress, we secured two modest grants. The first, amounting to $8,000, was provided by Thies University. More recently, we received a substantial $100,000 grant from the Bill and Melinda Gates Foundation, specifically earmarked for data collection.

We are now poised to transition from the research phase to commercial production. By the end of this year, we plan to launch the first version of our product to the market, marking a significant milestone in bringing our research to practical application.