Infectious diseases continue to be a major public health concern, particularly in developing nations like India where they are the leading causes of mortality, morbidity, and disability. The World Health Organization (WHO) estimates that India contributes to more than a quarter of the world’s 10 million tuberculosis (TB) cases.[1] Close to 505,000 people died from tuberculosis in 2021 in India, a substantial 10% increase from the previous year[2]. Delayed diagnosis, lack of rapid, sensitive, accurate and portable diagnostics, manual errors in interpretation of tests, insufficient training of workforce to use these advanced tests, and expense incurred in the procurement & maintenance of these equipments are a few challenges in the current TB diagnostics spectrum that contribute significantly to India’s outsized TB burden.
Early diagnosis of TB could be a game-changer that can save millions of lives. The WHO estimates that each year, approximately 3 million cases of TB are missed or go untreated globally. [3] These “missing millions” of TB-affected people who are neither detected nor treated, result in worse outcomes, more TB exposures[4] and may result in continued transmission of this potentially fatal disease by acting as community reservoirs of the TB bacterium. Part of the reason for the missing millions are screening and diagnostic gaps in the current TB care cascade. Ranging from the plight of an individual who must wait for weeks to get a laboratory report, to individuals who may not have access to TB diagnostic tests, to people who may not seek care owing to the socio-demographic variables and myths around TB, to those patients who fail to be diagnosed despite reaching health facilities due to lack of diagnostic facilities or the absence of trained laboratory technicians and pulmonologists to detect and treat the condition.
The efficacy of culture-based TB diagnosis, which has been considered the gold standard[5] in clinical practice, is hindered by high test turnaround times. To address this, MTB/RIF assays like GeneXpert and Truenat[6] which are semiautomated rapid molecular testing methods for identifying Mycobacterium tuberculosis DNA and resistance to the drug rifampicin were developed. However, as of 2022, their application continues to be far too restricted with only 1.9 million (33%) people taking it as an initial diagnostic test worldwide.[7] Although GeneXpert and Truenat still remain the primary testing mechanisms for TB, the healthcare sector is proactively looking for alternatives to address these issues of limited test menus, high cost, low throughput, sample type limitations, and dependencies such as electricity and trained human personnel. To fill these gaps and improve diagnostic competencies, more recent technologies are being developed.
One that holds tremendous potential of transforming the infectious disease diagnosis landscape is artificial intelligence (AI) along with other adjunct technologies like machine learning. Researchers and innovators are utilizing the power of AI to supplement and enhance the current diagnostic methods in recognition of the need for increased accessibility, cost-effectiveness, sensitivity, awareness, and infrastructure. Who would have thought that TB diagnosis could be done within minutes using artificial intelligence by analyzing a chest X-Ray with a smartphone? Or that one can diagnose a condition like TB using sound analytics? But both are a reality now! AI algorithms are now being trained to analyze medical images, cough patterns, and vocal tones, with a level of accuracy that is comparable to that of human radiologists.[8]
Enhanced Accuracy: AI algorithms can analyze massive amounts of medical data, such as radiographic scans and clinical records, with high precision by learning from large past datasets using machine learning techniques, allowing them to discover trends and detect TB-related anomalies in diagnostic images more correctly over time. In clinical trials, the pooled sensitivity and specificity of AI methods in medical imaging for Pulmonary TB were estimated to be 91% and 65%, respectively.[9]
Improved Efficiency: AI can streamline the diagnostic process by automating tasks such as image analysis and interpretation, reducing the workload on healthcare professionals and facilitating faster diagnosis. Computer-Aided Diagnosis (CAD) systems for chest radiographs using AI have recently shown great efficiency as a second opinion for radiologists.[10]
Faster TB detection rates: AI can aid in faster medical diagnosis by analyzing medical images such as chest X-rays accurately within a few minutes. For example, an AI-based software qXR rapidly classifies X-ray scans, identifies lung abnormalities, and highlights them on the X-ray, enabling the detection of TB within minutes, and linkage to treatment on the same day.[11] Fast detection rates of AI-based tools can lead to significant time savings, allowing healthcare providers to allocate more time to patient care and treatment planning.
Cost Effectiveness: AI algorithms can handle an immense quantity of data e.g. chest X-ray scans effectively, cutting down on time and operational expenses. This is accomplished by automating, enabling remote monitoring & optimizing the diagnosis process. The use of sophisticated and expensive therapies is also decreased by AI’s improved early recognition of indications and trends.[12]
Accessible Diagnosis: The easy availability of telecom infrastructure, free web tools, and cost-free host servers contribute significantly to the accessibility of AI technology for TB diagnostics. These components are now making it possible to leverage remote access, user-friendly interfaces, and affordable computing and storage resources, democratizing the use of AI and boosting TB diagnostic capabilities, especially in areas with limited resources. For example, an AI-based, easily accessible mobile platform “Swaasa” which is currently being validated with support from India Health Fund & ACT Grants, records cough sounds from a patient using a mobile phone’s microphone, and analyzes them to decode unique cough signatures, to detect the possible presence of TB in just a few seconds.[13]
Poor Data Quality: AI algorithms require large and diverse datasets for training and validation before they can be used in clinical practice. However, in the case of TB diagnosis, high-quality annotated data may be scarce or insufficient. [14] Hence, ensuring good-quality datasets will be essential for AI integration.
Data unavailability: A lack of comprehensive data reflecting all socio-economic categories can jeopardize the reliability of judgements for underrepresented groups.[15] Unfortunately, TB diagnosis requires a combination of clinical, radiological, and laboratory tests, which can be expensive and time-consuming and requires interoperable solutions. Hence, we need to work towards making comprehensive quality data available through the right platforms to make the best out of AI.
Regulatory Considerations: Integrating AI into TB diagnosis raises important ethical and legal concerns. Patient privacy, data security, and informed consent need to be carefully addressed to protect sensitive health information and ensure compliance with existing regulations. Additionally, algorithmic transparency and accountability should be prioritized to maintain trust and mitigate the risk of biased or discriminatory outcomes.
Human-AI Collaboration: Successful integration of AI in TB diagnosis necessitates effective collaboration between AI systems and healthcare professionals. For example, interpreting chest X-rays can be challenging & subjective, even for experienced radiologists. Clinicians need to understand the capabilities and limitations of AI tools to interpret their outputs accurately and make informed diagnostic decisions as the final clinical decision lies with the expert clinician and AI can assist them to make their work simpler. Additionally, AI algorithms [16]need to be integrated into existing clinical workflows, which can be complex and time-consuming. It also requires training[17] and adapting healthcare providers to effectively tap into & utilize AI technologies.
Lack of technical and financial resources: Implementing AI solutions for TB diagnosis requires technical expertise, infrastructure, and financial resources.[18] Many healthcare settings, particularly in low-resource areas, may face challenges in procuring and maintaining the necessary hardware, software, and connectivity required for AI integration. So, optimizing the available resources in AI will be necessary.
In conclusion, the incorporation of AI into TB diagnosis provides a significant opportunity to automate and improvise TB screening and detection. However, to enable successful adoption, challenges such as formulating standardized reporting guidelines, applying diagnostic heterogeneity, sorting implementation issues, strengthening technical know-how, developing AI technological assessment frameworks, and boosting financial resources will all be needed. And for these, physicians, researchers, innovators, funders, implementers, and policymakers must work proactively and collaboratively.
Dr Aishwarya is a public health professional who is currently working as an Associate: Portfolio & Programs at India Health Fund. She is a physiotherapy graduate from Seth GS Medical College & KEM Hospital, Mumbai & a Master’s in Public Health in Health Administration from the Tata Institute of Social Sciences, Mumbai. She has worked in various domains of healthcare such as physiotherapy, program implementation, research & community engagement.
Views expressed are personal.
India Health Fund is registered as Confluence for Health Action and Transformation Foundation (CHATF), a Section 8
charitable company incorporated in India, supported by the Tata Trusts.