Atlas of TB Innovations: Mapping the Future

LPA Ai Reader

LPA Ai Reader

LPA AI Reader—the AI‑powered solution developed by Wadhwani AI to automate the reading, interpretation, and transmission of LPA test results in the diagnosis of tuberculosis (TB).

Purpose & Impact

The LPA AI Reader is built to automate the interpretation of Line Probe Assay (LPA) results—used widely to detect drug-resistant tuberculosis (DR‑TB). Traditionally, lab technicians manually interpret test strips, which is laborious, time-consuming (up to an hour for a sheet of 24 strips), and prone to human error.

Key Objectives:

1. Speed up interpretation and reduce turnaround times.
2. Minimize errors in human interpretation of LPA strips.
3. Free up valuable lab technician time for other critical tasks.

Rationale:

1. India has one of the highest burdens of DR‑TB globally, with manual processes historically taking up to three months for diagnosis—reduced to just 5 days with LPA, yet still labor-intensive.
2. Automating LPA interpretation can significantly scale diagnostic capacity, reduce errors, and speed up initiation of treatment to align key goals with India’s National TB Elimination Programme (NTEP) and its targets.

Advantages:

1. Reduce delays involved in the interpretation and sharing of results with end-users. Aids more prompt treatment initiation.
2. Standardize LPA sheet and patient result interpretation: Establishes a clear standard for the interpretation of LPA sheets and results through the solution.
3. Eliminate reporting and transcription errors: Eliminate errors associated with the manual entry of LPA results into designated MIS under NTEP.

Technical Foundations & Innovations

1. Automated LPA Interpretation via Computer Vision & ML
2. The system uses computer vision techniques to isolate individual LPA strips from a scanned sheet. This includes extracting each strip from possibly distorted or crumpled images.
3. It then applies image analysis—often edge detection—to pinpoint activated (dark) vs. inactive (light) bands on each strip.
4. A rule-based classification maps these band patterns to specific drug resistance profiles, effectively flagging drug sensitivity for TB strains.
5. To reduce manual burden, the system includes Optical Character Recognition (OCR models) to capture and log test details directly, cutting down on data entry errors.
6. Additionally, the AI tool also flags unclear or novel patterns for manual review by experienced technicians—helpful for both trust and safety.
7. High Reliability & Dataset Rigor
8. Developed with an extensive, diverse dataset of annotated LPA strips—labeled by microbiologists across India—the system achieves over 97% accuracy for all drugs tested in both first-line and second-line LPA tests.
9. Additionally, the tool is independently evaluated by the Indian Council of Medical Research (ICMR), wherein the tool demonstrated an accuracy of 97% for first-line and 100% for second-line LPA, which again vouch for the reliability of the developed AI tool

Workflow & Human-in-the-Loop Approach

1. Beyond purely technological innovation, Wadhwani AI emphasizes building a human-in-the-loop process:
2. AI handles the heavy lifting—locating strips, reading bands, and applying rules.
3. Ambiguous cases or low-confidence interpretations are automatically flagged for expert scrutiny—ensuring oversight and trust.
4. This approach also aids in integrating the tool into existing lab processes, addressing concerns around accountability, workflow changes, and infrastructure readiness.

How It Works: Three-Step AI Pipeline

Wadhwani AI structured the processing into three distinct stages:

1. Locating Test Strips
2. Detects printed box locations (where strips are placed) using template matching and spatial structure (assuming equal spacing among 24 strips).
3. Identifies the presence/absence of a strip at each location.
4. Reading Activated Bands
5. Once located, it applies edge detection algorithms on each strip to detect the presence (dark bands) or absence (light/empty) of specific markers.
6. Mapping to TB Variants
7. Based on the pattern of activated bands, the system applies deterministic decision rules (similar to those used by technicians) to classify the type of TB drug resistance and catch preparation errors.

This clever pipeline enables a reliable and explainable AI workflow to be embedded within existing lab processes.

Deployment & Human‑in‑the‑Loop Integration

1. The AI tool was conceptualized and developed in the context of the ongoing collaboration under the MoU, Ministry of Health and Family Welfare, Government of India.
2. Despite promising proof-of-concept results, real-world deployment brings challenges—such as integrating into lab workflows, ensuring accountability, infrastructure constraints, and building user trust through explainability and uncertainty metrics.
3. Wadhwani AI advocates a human‑in‑the‑loop approach where lab technicians validate AI interpretations, blending efficiency with safety and confidence.

The Way Forward

With the LPA integrated into the Ni-kshay platform, the next phase focuses on scaling adoption nationwide through trainings and capacity-building workshops under the guidance of and in coordination with the Central TB Division and Ministry of Health and Family Welfare to empower IRL and CDST labs across India.

Potential Impact

1. Deployment across 97 NTEP-certified Culture DST and IRL labs.
2. Improved reading, interpretation, transcription, and digitization of LPA results,
3. reducing overall time to 15 min.
4. Overall reduction in turnaround times of up to 12 hours for every patient tested.
5. Faster treatment initiation, with an estimated 200,000 patient-days saved annually.
6. An estimated 12,000 additional patients benefited per year from receiving the correct treatment regimen.

Summary Table

Genki

Genki is an AI-powered lung health screening tool developed by DeepTek. Genki leverages advanced deep learning techniques, including attention mechanisms, which help the model focus on the most critical regions of chest X-rays for accurate TB detection. It includes modules for domain shift detection (to account for variations in imaging sources, devices, and settings) and foreign-object detection (to identify interfering artifacts like pacemakers or leads) for more reliable performance in diverse real-world conditions.

1. Deployment Options
2. On premise: (Edge Deployment) Genki runs on local hardware—such as laptops or mobile units—with the ability to analyze X-rays within 20–30 seconds, even offline.
3. On cloud: (Cloud Integration) When connectivity is available, data synchronizes with the cloud for centralized processing, reporting, analytics, and program oversight.
4. AI + Human Collaboration (Expert‑in‑the‑Loop)
5. The system initially pre-screens images using AI, then radiologists review, validate, or modify the outputs. This expert feedback is used to continuously retrain and enhance the model, reducing errors over time.
6. Workflow & Integration
7. RIS‑VIM Module (Radiology Information System-Vulnerability Index Management): Manages radiology data and vulnerability indexing.
8. Augmento (Cloud Platform): Handles centralized collaboration, reporting, and analytics dashboards and integrates with hospital systems like PACS, RIS, HIS, and national TB databases (e.g., Nikshay in India).
9. Smart Notifications & Analytics: Provides automated reminders, follow-up scheduling, and operational dashboards for program management.
10. Performance & Productivity Gains:
11. In internal evaluations, Genki showed, sensitivity of ~93%, specificity of ~83%, and an AUROC of ~93%. Clinically, it improved radiologist productivity by 86% and reduced turnaround time by 46%. Field deployments (e.g., in Chennai) achieved 98% sensitivity and 96.9% accuracy in real-world screenings.
12. Regulatory & Global Readiness
13. Certifications: US FDA 510(k) cleared, as well as approvals from Indian CDSCO, Thai FDA, Kenya Board, Singapore HSA, and EU MDR, and recommendation by WHO for TB CAD software.
14. Deployment Reach: Used across 800+ screening sites globally, processing over 2 million individuals, with successful deployments in India, the Philippines, Thailand, Mongolia, Kenya, and more.

Accuracy & Performance

1. AUROC (Area Under Receiver Operating Curve): 0.93 [95% CI: 0.92, 0.95]
2. Sensitivity: 0.93 [95% CI: 0.88, 0.97]
3. Specificity: 0.83 [95% CI: 0.82, 0.83]

Real-World Use Cases

1. Mobile screening vans in India, Africa, Southeast Asia
2. Prison screenings and migrant camps
3. Community TB campaigns by public health departments
4. Augmenting radiology teams in high-burden hospitals

Reference:

1. Codlin, A.J., Dao, T.P., Vo, L.N.Q. et al. Independent evaluation of 12 artificial intelligence solutions for the detection of tuberculosis. Sci Rep 11, 23895 (2021). https://doi.org/10.1038/s41598-021-03265-0
2. Jayaraman P, S S, Paul S, Pant R, Gupte T, Kulkarni V, Kharat A. Artificial intelligence as a proficient tool in detecting pulmonary tuberculosis in massive population screening programs: a case study in Chennai, India. J Rural Med. 2025 Jan;20(1):13-19. doi: 10.2185/jrm.2024-015. Epub 2025 Jan 1. PMID: 39781302; PMCID: PMC11704598.

ProDigi Health Platform for TB screening with PRORAD ATLAS ULTRAPORTABLE Digital X-Ray system

The PRORAD Atlas Ultraportable X‑Ray System is a cutting-edge, backpack-style radiography solution designed for definitive imaging anytime, anywhere—from remote health camps to bedside diagnostics.

Key Highlights

1. Ultra-Lightweight & Portable: Weighing approximately 2.80 kg (including battery), this featherweight unit offers exceptional mobility.
2. Rapid Setup — “Aim and Expose”: Engineered for efficiency with intuitive ergonomics for field use and quick patient imaging.
3. Smart, User-Friendly Control Interface: Features a digital or touchscreen display to set exposure parameters with ease.
4. Extended Battery Life: Equipped with a rechargeable battery (11.1 VDC), it ensures prolonged field usage.
5. Ultra-Low Radiation Dose: Delivers exceptional image quality while maintaining a low radiation exposure profile, thanks to excellent MTF and DQE performance.
6. Seamless DICOM Integration: Designed for streamlined image handling and connectivity with health IT systems.

Technical Specifications of PROGNOSYS MEDICAL SYSTEMS (PRORAD)

Regulatory Approval & Intended Use

AERB, CDSCO, ICMR recommended, BIS, BAPTEN (Indonesia), FDA Philippines, US FDA Approved

Why It Stands Out

1. Unmatched Mobility: Its compact size and light weight enable healthcare professionals to deliver imaging services virtually anywhere.
2. High-Quality Imaging at Low Dose: It ensures both diagnostic accuracy and radiation safety, making it ideal for sensitive or mobile setups.
3. Workflow-Friendly: Touchscreen controls, DICOM compatibility, and rapid setup make it ideal for urgent care and ergonomically demanding situations.
4. Versatile Applications: Perfect for home healthcare, remote screening, bedside diagnostics, and emergency deployments.

The PRORAD Atlas Ultraportable X‑Ray System is the perfect blend of portability, performance, and ease-of-use. With a microprocessor-controlled high-frequency generator, ergonomic design, low radiation dose, and seamless digital workflows, it's ideally suited for modern medical scenarios that demand mobility and reliability.

PRORAD ATLAS ULTRAPORTABLE digital x-ray system is a state-of-the-art ultra-portable solution that comprises of high frequency x-ray generator that is microprocessor-controlled for precise x-ray delivery, an ergonomically and aesthetically designed unit with an LCD touch screen display of x-ray parameters and APR settings. It is a compact and lightweight X-ray imaging device that is designed for use in a variety of settings, including emergency rooms, intensive care units, and remote or mobile clinical settings.

PathoDetect PCR Kits with Resistance Detection

The PathoDetect MTB RIF & INH Resistance Detection Kit is a robust, indigenous RT-PCR that offers fast, sensitive, and comprehensive TB diagnosis—including drug resistance profiling in under 2 hours. It consolidates tests into a single, streamlined assay. It's globally aligned yet tailored for India’s specific public health needs, with minimal infrastructure requirements and deep integration into national TB control systems.

Features of PathoDetect TB Kits

1. Fully Integrated Sample-to-Result Workflow
2. The PathoDetect MTB RIF & INH Resistance Detection Kit is designed for use with the Mylab Compact™ device, offering fully automated testing from sample loading to result generation in under 2 hours.
3. This represents a closed NAAT system, where steps like extraction, amplification, and detection occur within a sealed, self-contained environment—minimizing manual intervention and contamination risk.
4. Automated, High-Throughput Platform
5. Supports parallel processing—handling multiple samples in one run (typically up to 16) and delivering robust contamination control, ease-of-use, and scalability.
6. By combining the cartridge-based kit with an automated PCR platform, the workflow is streamlined and operator-independent, a hallmark of closed-system designs.
7. Ambient-Friendly Design:
8. Operates effectively without cold storage (unlike traditional cold-chain-dependent PCR kits), suited for rural, mobile, or low-infrastructure setups
9. Multiple TB Testing Modalities
10. MTB Detection Kit: Rapidly identifies Mycobacterium tuberculosis using multicopy sequence IS6110 for enhanced sensitivity.
11. MTB/NTM Discrimination Kit: Differentiates MTB from non-tuberculous mycobacteria (NTM).
12. MTB RIF & INH Resistance Detection Kit: Simultaneously detects TB and key drug-resistance markers for rifampicin and isoniazid.
13. All kits emphasize high sensitivity, low contamination risk, quality control, and fast results as system hallmarks.
14. Regulatory Approval & Field Validation
15. The MTB RIF & INH kit is India’s first made-in-India TB detection kit with CDSCO, TB Expert Committee, and ICMR approvals and has gone through multicenter clinical evaluation, reinforcing reliability and performance.

The PathoDetect TB kits, especially the MTB RIF & INH Resistance Detection Kit, function as a closed NAAT system—offering a sealed, automated, and efficient workflow with rapid (approx. 2-hour) results and strong regulatory backing. These features facilitate deployment even in semi-automated or near-patient settings, supporting India’s TB elimination goals.