Union Conference 2024: Versatile technology driving tuberculosis elimination efforts across Africa through active case finding

Breakthrough research presented at The Union World Conference on Lung Health demonstrated the role of innovative technology in driving active case finding efforts, slowing the progression of tuberculosis (TB) across African nations

• AI and computer-aided detection (CAD) is driving the discovery of new TB cases across Africa
• New cases uncovered indicate incidence rates of TB almost seven times higher than national estimates
• Experts call for greater TB surveillance and wider use of new screening technologies in high-burden communities

Bali, Indonesia; November 15, 2024: Tuberculosis (TB) experts have revealed how new technology is enhancing active case finding (ACF) efforts in communities across Africa.

Research from Ethiopia, Kenya, and Nigeria has demonstrated how artificial intelligence (AI) and computer-aided detection (CAD) software are supporting increased detection of the disease at The Union World Conference on Lung Health.

Experts from REACH Ethiopia presented new findings on how AI-powered TB screening is being used across urban slum, rural, and pastoralist communities to detect TB across the country. Over 12,000 people were screened using a chest X-ray supported by AI detection, with individuals either presenting with TB-suggestive symptoms or X-ray findings suggestive of TB then invited for bacteriological testing.

TB-suggestive symptoms were detected in 8.5% of community members screened, with the TB incidence rate found to be 927 per 100,000 population – nearly seven times higher than the national estimation.

The researchers concluded that AI-assisted X-ray screening exhibited notable sensitivity, surpassing symptom-based screening and significantly contributing to case detection rates in Ethiopia.

Experts from KNCV Nigeria presented separate research on how AI-aided screening of non-symptomatic groups is helping to close the TB case finding gap – with their study results indicating that surveillance for TB among high-burden populations should be increased.

Data from AI-supported scans of nearly 26,000 individuals from across different communities were retrospectively examined using an AI-aided portable digital X-ray tool. Presumptive cases were sent for further examination using GeneXpert, with radiologists also interpretating scans to make clinical diagnoses.

The researchers concluded that AI is playing a significant role in TB programming by helping to detect subtle, subclinical lung lesions earlier – which likely would have otherwise remained undiagnosed.

Researchers based in Kenya showed how eight digital chest X-ray tools equipped with CAD software could provide a cost-effective strategy for early active TB case identification – particularly in high-burden countries such as Kenya. The CAD-equipped X-ray tools were used as part of the ‘Introducing New Tools’ project, funded by USAID.

The tools were incorporated into Kenya’s TB targeted outreach programme with screening algorithms and alongside symptom screening for all outreach attendees.

Nearly 16,000 individuals were reached through the programme, which streamlined the process for identifying people with TB whose symptoms required further laboratory investigation. The study also found a high TB positivity rate among individuals with elevated CAD scores, stressing the efficacy of community-based screening efforts using CAD-supported tools in early disease detection.

The researchers concluded that scaling up the deployment of this technology holds promise for significantly enhancing TB detection rates in Kenya – demonstrating the important role that the strategic deployment of innovative tech solutions could play in eliminating the disease.

Lead researcher Rhoda Karisa said: “Our findings highlight the transformative potential of using AI-enabled digital chest X-ray tools in community-based TB screening programmes.

“By leveraging this innovation, countries can enhance early detection of TB and ultimately move a step closer to the goal of eliminating TB in high-burden countries like Kenya.”

Speaking at The Union World Conference on Lung Health in Bali, Dr Cassandra Kelly-Cirino, Executive Director, International Union Against Tuberculosis and Lung Disease (The Union) said: “Innovative technologies, such as AI and computer-aided detection, have the potential to change the landscape for detection of TB, allowing people with the disease to access treatment more quickly and helping to break the chain of transmission.

“However, innovation in healthcare is only as valuable as the number of people it reaches. To fully realise the potential of AI and CAD in TB detection, we must go further in using them as widely as possible to help make the elimination of TB a reality. Research is proving the efficacy of these tools – it’s up to us all to take action in funding and implementing them worldwide.”

Note: Full abstracts below

ENDS

Notes to editors:

The Union

Established in 1920 as the world’s first global health organisation, the International Union Against Tuberculosis and Lung Disease is committed to eradicating tuberculosis and lung disease, leading to a healthier world for all. Its members, staff, and consultants work in more than 140 countries globally.

The Union is a global membership, technical and scientific organization, striving to end suffering due to tuberculosis and lung diseases by advancing better prevention and care. We seek to achieve this by the generation, dissemination and implementation of knowledge into policy and practice.

The Union’s approach to tackling global health problems is unique – KNOW. SHARE. ACT. We start with developing knowledge through global research (KNOW), which we then share as widely as possible (SHARE) and turn that into the real action to save lives on a local level (ACT).

The Union’s work is exemplified by its core values of quality, transparency, accountability, respect, and independence.

The Union social media

Twitter: @TheUnion_TBLH
Facebook: The Union Lung Health
LinkedIn: International Union Against Tuberculosis and Lung Disease (The Union)

ANNEX – Abstract Summaries 

Artificial intelligence for TB screening to improve active TB case detection among urban slum, rural and pastoralist TB-key affected and vulnerable populations in Ethiopia

Background: Limited access to X-ray facilities and a shortage of trained radiologists in resource-constrained settings like Ethiopia impede its widespread use. Recent advancements in Computer-Aided Detection and Artificial Intelligence (AI) offer promising alternatives for digital CXR analysis, particularly in TB screening and triage. This pragmatic evaluation aims to comprehensively assess a novel ultra-portable digital X-ray system with AI, gathering crucial data to elucidate its functionality and impact. Through systematic documentation and analysis, the study seeks to provide evidence ​for the program.

Design/Methods: This study involved 12,181 individuals from TB key affected and vulnerable groups across three diverse settings: urban slums, rural areas, and pastoralist communities, undergoing Chest-Xray and symptom-based screening. Onsite X-ray image interpretation, supported by Quire AI and by radiologists upon return, was conducted for all images. Individuals with TB suggestive symptoms and/or abnormal chest X-ray findings by AI interpretation and/or radiologist assessment were invited to provide specimen for bacteriological tests. Data was collected using a structured format designed for this purpose.

Results: A total of 12,181 individuals (39% females, 61% males) underwent TB screening. Symptom screening identified TB suggestive symptoms in 8.2% (95% CI 7.7-8.7%) of community members. The concordance rate between AI X-ray interpretation and clinical symptom screening was 88%, while a 95% concordance rate was observed between AI interpretation and radiologist readings. Among 802 X-ray images with abnormal findings, 98% exhibited active TB indicators. The TB incidence rate was 927 per 100,000 population (95% CI 757.4-1,098), nearly 6.7 times higher than the national estimation. The sensitivity rate of AI-assisted X-ray screening was 93.5% (95% CI 86.5-96.9%) with a specificity of 63.7% (95% CI 61.0-66.3%). The tool demonstrated a high negative predictive value (99.25, 95% CI 98.37-99.66).

Conclusions: AI-assisted X-ray screening exhibited notable sensitivity, surpassing symptom-based screening, and significantly contributing to case detection rates.

Closing the TB case finding gap through artificial intelligence (AI)-aided screening of non-symptomatic population: Katsina State experience

Background and challenges to implementation: Missing TB cases have continued to be the bane of TB programming in Nigeria even though there is a considerable improvement in case notifications in recent time. To address this challenge, intensive screening is directed to carefully selected communities wherein dwellers are massively screened, regardless of their symptoms. It is however fascinating to see that apparently healthy individuals with no symptoms but high index of suspicion from AI detected lesion turned out to be TB cases. This study therefore aimed to estimate the proportion of non-symptomatic patients whose eventual diagnosis was enhanced by artificial intelligence and hence help close the TB case notification gap in Katsina State Nigeria.

Intervention or response: A cross-sectional retrospective review of data from AI-enabled portable digital Chest X-ray (CXR) between January 2022- January 2024 was conducted. Programmatically, CAD4TB was set at 0.50 (50%) for presumptive TB threshold. Community members were screened by AI-aided Portable Digital Xray, and their score read immediately. Presumptive cases were sent for GeneXpert. If unable to produce sputum, such film was sent to qualified radiologists for interpretation and possible clinical diagnosis using XMAP.

Results/Impact: 25,993 people were screened in different communities, 659 were without any TB symptom. This non-symptomatic population yielded 39 presumptive TB detected by CAD4TB at 0.5 score. This resulted in the diagnosis of 11 TB cases including 1 bacteriologically diagnosed.50 score. This resulted in the diagnosis of 11 TB cases including 1 bacteriologically diagnosed.
 

Conclusions: Just as evidenced in other spheres, artificial intelligence is playing a big role in TB programming by helping in early detection of subtle, subclinical lung lesions which would have otherwise remained undiagnosed. There is a need to intensify TB surveillance among susceptible populations regardless of TB symptom, possibly a need to scale up to national TB program screening policy.

Enhancing TB detection through the integration of computer-aided detection (CAD) software in chest radiography: Findings from targeted outreach programs in Kenya

Background: Integration of Chest radiography (CXR) with computer-aided detection (CAD) software has been endorsed by the World Health Organization (WHO) as a pivotal screening tool for tuberculosis (TB) detection. This study investigates the utilization of CXR with CAD software within targeted outreach programs as a cost-effective strategy for early active TB case identification, particularly in high burden countries like Kenya.

Design/Methods: Kenya implemented the use of 8 digital chest X-ray machines equipped with CAD software as part of the Introducing New Tools USAID funded project (iNTP). These machines were seamlessly integrated into Kenya's TB targeted outreach program, with screening algorithms devised to incorporate CXR alongside symptom screening for all outreach attendees. Quarterly outreaches were conducted across eight designated sites.

Results: A total of 15,916 individuals were reached through the screening outreach initiative in 2022 and 2023, with 55% (8,680) being male and 56% (8,957) presenting with TB symptoms. Analysis of CAD-generated scores revealed that 5% (793) scored above 60, while an additional 5% fell within the 40-59 score range. Among those with a CAD score above 60, 72% (572) underwent laboratory investigations for TB confirmation, compared to 57% (476) for the 40-60 score range. Notably, 28% (163) of individuals with a CAD score above 60 tested positive for TB through laboratory investigation, including 47 patients who were asymptomatic. The TB positivity rate among individuals with a CAD-score of 40-60 was 6%.

Conclusions: The implementation of CXR with CAD facilitated a streamlined process for identifying individuals requiring further laboratory investigation, thereby reducing unnecessary Gene-Xpert testing during outreaches. Moreover, the study found a high TB positivity rate among individuals with elevated CAD scores, underscoring the efficacy of community-based screening efforts augmented by AI-enabled X-rays in early disease detection. Scaling up the deployment of this technology holds promise for significantly enhancing TB detection rates in Kenya