TB Detection - AI & Auscultation

The problem we aim to address is the inefficiency and inaccuracy of tuberculosis (TB) screening, which contributes to a substantial diagnosis gap and missed opportunities for early intervention. TB remains a significant global public health challenge. In 2023, TB reclaimed its spot as the most deathly infectious globally, claiming nearly 1.6 million lives annually despite being a treatable disease. The burden of TB is particularly profound in South Africa, with an estimated 304,000 infections in 2021 and a staggering figure of 55,000 annual TB-related deaths in the country. The financial strain on healthcare systems is evident, with roughly US$210 Million allocated specifically for TB management in South Africa in 2022. The shortcomings of the current screening method exacerbate this burden, as it erroneously excludes 29% to 44% of TB-positive individuals from the outset and subjects 36% of TB-negative cases to unnecessary costly confirmation tests. To provide context, in 2021, only 172,194 cases were recorded for treatment out of an estimated 304,000 infections.

Globally, an estimated 10.6 million individuals fall ill from TB annually, with approximately 4.2 million TB patients missed each year. This disparity highlights the urgent need for improved TB screening methods to identify and treat cases early.  The lack of access to accurate and efficient TB screening exacerbates the diagnosis gap, resulting in increased mortality and economic costs. According to the UN, failure to address these issues is expected to result in an additional 43 million people developing TB and resulting in economic costs totalling US$1 trillion by 2023.

The solution is an Artificial Intelligence(AI)-enabled digital stethoscope designed to detect Tuberculosis (TB). It consists of three main parts: (i) The Stethoscope Hardware and supporting Firmware, (ii) The Application software and supporting Cloud Infrastructure, and (iii) a TB detection AI Model. The Stethoscope connects to the Application via Bluetooth. The Application guides the user through the chest auscultation procedure, indicating where the nurse/user should place the Stethoscope while guiding the patient's breath rate. The Stethoscope records the patient's lung sounds and background noise and sends the data to the Application. The Application transfers the data to the locally run AI container (for offline usage, a cloud-based version is also available). The AI container first checks the quality of the recorded data. If background noise exceeds a threshold, the user will be requested to repeat the recording. The lung audio recordings and clinical information (HIV status, etc.) are passed to the TB AI model. The AI Model then processes the data and outputs a TB prediction within seconds. The AI container returns the TB prediction to the Application, which is interpreted into an instruction displayed to the user. This instruction can be tailored to the use case and operating environment.

AI Diagnostics has developed the largest TB-validated auscultation database in the world (>30,000 validated recordings), which serves as the training data for the AI Model. The training database was initially collected in a local clinical study that included 29 South African sites. The primary advantages of the solution lie in its enhanced detection accuracy, with sensitivity and specificity rates of 83% and 76%, respectively, as opposed to 71% and 64% for the current 4-symptom questionnaire. This translates to a significant reduction in missed TB cases (41%) and potentially lowers overall TB diagnosis costs by 33% through fewer unnecessary confirmation tests. The solution’s customizability offers flexibility in configuring decision thresholds based on specific environments or applications. The consumable-free nature of the tool eliminates the need for disposable components, addressing logistical challenges associated with storage, distribution, and safe disposal of consumables. Furthermore, the digitization of screening events in real-time provides valuable epidemiological insights, facilitating a more informed approach to TB diagnosis and management.

This solution is currently being reviewed by the World Health Organization TB advisors in an international study across 3 continents. Using the same technology, AI Diagnostics has plans to expand the detection capabilities into more pulmonary, cardiovascular, and abdominal diseases.

The target population for the solution is individuals living in resource-limited communities with high burdens of tuberculosis (TB) around the world. These communities often face significant challenges in accessing timely and accurate TB diagnosis due to limitations in traditional diagnostic methods, such as the low sensitivity and specificity of the symptom-based questionnaire, the lack of infrastructure and trained personnel. Currently, vulnerable populations within these communities, including those with limited access to healthcare facilities or living in remote areas, are underserved in terms of TB diagnosis. They experience delays in diagnosis, leading to prolonged illness, increased transmission of TB within the community, and higher mortality rates.

Our AI-enabled Digital Stethoscope addresses these needs in several ways. Firstly, by eliminating the need for complex equipment, it makes TB screening more accessible, even in areas with limited infrastructure. This empowers healthcare workers, including those with less experience, to conduct reliable TB screenings, thus improving access to diagnosis for underserved populations. Early diagnosis reduces patient morbidity, it improves treatment outcomes and reduces period contagiousness, reducing the chance of spread amounts family members and peers.

The benefits of the solution extend beyond the individual. The digital stethoscope is designed to be cost-effective, reducing unnecessary expensive bacteriological confirmation tests. This translates to significant cost savings for resource-constrained healthcare systems. Additionally, the digitized screening data collected by the digital stethoscope provides invaluable insights into TB prevalence and patterns within these communities.  Armed with this information, healthcare officials can make data-driven decisions for TB control and resource allocation.  The AI stethoscope offers a promising solution to bridge the diagnostic gap and improve the lives of millions living under the shadow of TB. 

Our team is based in South Africa, a country with an extremely high burden of TB. This geographical proximity provides us with a firsthand understanding of the socio-economic and healthcare landscape in the region. Being embedded within the community allows us to empathize with the challenges faced by individuals affected by TB on a daily basis. We understand the nuances of the local context, including language preferences, cultural sensitivities, and healthcare-seeking behaviours.

We actively engaged with community members, healthcare providers, and stakeholders throughout the development process to ensure that our solution addresses their specific needs and priorities. For example, our enrollment model leveraged 29 different sites strategically located within the community to maximize participation. During our studies, we employed recruiters and promoters who were familiar with the local culture and language. Individual connections have been made and preserved at each regulatory level to obtain regulatory approval. 

The founders consist of a Mechatronic engineer, an electrical engineer and a chemical engineer turned data scientist, providing a range of expertise to oversee product development. The team has 20 years of medical device development experience. The team is supported by a regulatory advisor, business advisor, and clinical advisors to fill blind spots. The team has partnered with local manufacturers to produce the devices and connected with sales networks from which to sell the product. The team is well-balanced and suited to deliver on AI Diagnostics' immediate goals.

Although the solution is pre-commercialisation, it has been launched across 3 continents, with 20+ users across different demographics. The project has received raised pre-seed investment funding which was used to take the prototype to pilot.

At AI Diagnostics we want to serve humanity by being the difference in the fight against TB GLOBALLY. We are well positioned to do this in South Africa but know that operating in other jurisdictions comes with issues only overcome through partnerships. We believe MIT Solve provides the ideal platform and network to form these key partnerships which will allow us for the next torch bearer to put a stop to this prolific but curable disease. 

AI-enabled stethoscopes have been used to detect diseases that physicians can typically detect. In these cases, particular sounds are annotated as being indicative of the disease or not, and the model is trained to identify such sounds. In contrast, detecting TB using a stethoscope is truly groundbreaking in that it is NOT accurately diagnosed amongst experienced physicians, and prior to our study the sounds indicative of TB were not well understood by the clinical world. During our training, the AI model was only provided the bacteriological TB test results and with our custom architecture, we allowed the model to converge on what was previously indecipherable TB audio features.

The digital stethoscope's power extends beyond raw audio analysis. The model also integrates clinical data about the patient. This additional information allows the AI to dynamically adjust its interpretation of lung sounds. By factoring in age, medical history, and symptoms, the AI achieves a more nuanced understanding of the patient's condition, further enhancing diagnostic accuracy. 

TB is remarkably socioeconomically correlated, preying on poor congested communities, where malnutrition and HIV infection are prevalent. The infectious nature of TB often leads to one or both breadwinners in households being put out of work for extended periods, further perpetuating the poverty spiral and enhancing the family's chances of contracting TB in the future. TB is also notoriously challenging to detect while the healthcare systems predominantly rely on symptom screening as the initial indicator of disease. Patients who are already symptomatic are also already contagious and thus have likely spread the disease within their communities.

At AI Diagnostics, we are addressing this major underlying cause by creating a low-cost and accurate TB screening tool that can detect TB in patients before they become symptomatic and contagious. The primary advantages of the solution lie in its enhanced detection accuracy, with sensitivity and specificity rates of 83% and 76%, respectively, as opposed to 71% and 64% for the current 4-symptom questionnaire. This translates to a significant reduction in missed TB cases (41%) and potentially lowers overall TB diagnosis costs by 33% through fewer unnecessary confirmation tests. 

Early detection allows infected patients the opportunity to return to work within a week of treatment, putting a halt to the perpetuation poverty spiral. Early detection also reduces the risk of transmission to family members and peers reducing the continual spread of the disease. 

The WHO, UN StopTB partnership's Global Plan to End TB 2023-2030 and the South African National Strategic Plan 2023-2028 have all focused on accurate low-cost, portable TB screening as an essential component in ending TB.

Increased TB-infected patient notification to treatment.
This is a direct reflection of the product's detection sensitivity which has been a key metric in development and the result thereof will be tracked in postmarket impact studies. 

Decrease in unnecessary expensive time-consuming bacteriological confirmation tests. 
This is a direct reflection of the product's detection specificity which has been a key metric in development and the result thereof will be tracked in postmarket impact studies. 

Overall decrease in TB prevalence. 
Year incident rates of TB in South Africa have started to plateau at over 300,000 cases per year. The stopTB partnership has modelled the incident rate if the current status quo is maintained. Through early detection and therefore treatment, minimising transmission, AI diagnostic aims to ensure TB prevalence in subsequent years continues to fall below the StopTB estimates.

There are numerous supporting technologies in this solution which include; the Windows application and the cloud storage infrastructure, the Bluetooth-connected digital stethoscope and ambient noise sensor and audio pre-processing techniques. However, the core technology that powers AI Diagnostics' TB solution is the AI model which autonomously conducts the lung sound audio and clinical information interpretation making a TB infection prediction.

AI Diagntoics has created a proprietary model architecture, specialised in this application. The development utilises a unique assembly of existing technology blocks. By converting the audio signals into images using Fourier transforms AI Diagnostics is able to leverage the series of image recognition attention and perception functions typically synonymous with transformer models. AI Diagnostics also encodes clinical data (e.g. HIV status) into the actual audio data and then again introduces this clinical data into subsequent linear layer models, which collate the outputs of transformer models and clinical information into one final TB prediction. 

5 fulltime 

11 contract

20 collaborators - paid for though partnerships. 

Since January 2020
4.25 years

Our primary objective is to provide healthcare solutions to those who need it most, hence our slogan "Equity in Medical Care". As the leadership group we value peer diversity and the importance of having representation of the groups we intend to serve with in our team. AI Diagnostic currently has no hiring mandate specifically defining diversity criteria, yet the welcoming sentiment is very clear in the company culture and aligned with our documented vision towards contributing to an equitable world. 

Our customers are institutions that service the target population's healthcare needs. Customer contracts are currently managed on an ad-hoc basis, ranging from once-off all-inclusive contracts, per-use costs, to monthly license-free revenue models. 
To make it tangible the immediate customer personas and the reasons why they would be incentivised to adopt our solution are detailed below:

1. Internal occupational health units in the mining sector,

• Legal Bound - TB screening at least once per year as per the Mine Health and Safety Act, 1996.
• Litigation Avoidance - recently faced us$260 million settlement.
• Reduce transmission and employee absenteeism - Early detection.
• Social Pressure - TB innovation inclusion in their integrated report - SDG 3, 8 - boost investment opportunity.
• Social Pressure - TB prevalence in mines is 10 times the WHO threshold for a health emergency.

2. Private occupational health service providers,

• Competitive Advantage - Additional desirable service offering.
• High-Profit Potential - high patient throughput (100 patients a day) Charging R25 per employee can result in R55,000 in revenue and R44,000 in profit per unit monthly.
• Accelerator -> Patient medical scheme inclusion.

3. Non-governmental organizations (NGOs) offering TB screening services.

• Mandate alignment - 40 -> 60 % reduction in positives missed when compared to symptom-based screening. Early detection.
• Direct Cost Savings - reduced secondary bacteriological tests.
• Cost Effective - 6 times cheaper than X-ray-based alternative with equivalent detection accuracy.
• Active screening compatible (portable) -> Aligned with the 2023 - 2030 global Stop TB goals.

4. Private general practitioners,

• Competitive Advantage - Additional service offering.
• Attractive to Innovation Enthusiasts.
• Attractive to Impact Enthusiasts.
• Direct profit potential (Low) -> lower throughput -> Charging R50 per patient can result in R11,000 in revenue and R3,500 in profit per unit monthly.
• Accelerator -> Medical scheme inclusion.

5. Primary healthcare institutions in the private sector

• Competitive Advantage - Additional service offering.
• Direct profit potential (Medium) -> Medium throughput -> Charging R50 per patient can result in R22,000 in revenue and R14,500 in profit per unit monthly.
• Accelerator -> Medical scheme inclusion.

6. Internal occupational health units in the construction, manufacturing, and agriculture sectors

• Avoid litigation - due to employee and community neglect, and/or food processing complaints.
• Reduced transmission and employee absenteeism - Early detection.
• Accelerator -> Medical scheme inclusion.

7. Primary healthcare facilities in prisons,

• Litigation Avoidance - Legal requirement to provide inmates with adequate healthcare.
• Reduced transmission between inmates - Early detection.
• Social Pressure - 20 times the global prevalence.
• Direct Cost Savings - reduced secondary bacteriological tests.
• Cost Effective - 6 times cheaper than X-ray-based alternative with equivalent detection accuracy.

8. Public primary healthcare providers

• Direct Cost Savings - reduced secondary bacteriological tests.
• Cost Effective - 6 times cheaper than X-ray-based alternative with equivalent detection accuracy.
• Active screening compatible (portable) -> Aligned with the 2023 - 2030 global Stop TB goals & National Strategic Plan 2023-2028
• Indirect cost saving - Early detection -> reduced transmission -> reduced cases & severity -> reduced high care costs.
• Social Pressure - Greatest prevalence of the highly TB-burdened countries.

The development to date has been funded by Angel investors, Institutional VCs and National grants. 

AI Diagnostics is a for-profit company and the financial projections suggested a monthly break-even in Q2 of 2026, and a cumulative profit break-even Q1 of 2027. These projections are developed on a cost of service which is set at half of the direct savings customers gain from reducing the cost of downstream bacteriological TB tests alone. This allows institutions to pay for AI Diagnostics services out of their current budgets while reaping the essential benefit of early detection of TB-infected patients.