Page 105 - AI for Good-Innovate for Impact Final Report 2024

AI for Good-Innovate for Impact

complications and death. Aligning with SDG 3.3's focus on ending the TB epidemic. Also, early
treatment minimizes the duration of infectiousness, thereby curbing the spread of TB within
communities. This directly supports SDG 3.3's goal of controlling communicable diseases.

Finally, AI-based diagnosis has the potential to be deployed in resource-limited settings, 22-Medics
expanding access to accurate TB diagnosis in regions with limited healthcare infrastructure.
Matching with SDG 3.8's focus on achieving universal health coverage.

22�2�2 Future work

1� Data Collection: Collaborate with microbiology labs to establish standardized protocols
for capturing high-quality microscopic images of patient samples. This should ensure
consistency in the data used to train the AI model. Partner with research institutions and
public health organizations worldwide to create a comprehensive dataset encompassing
diverse geographical regions, patient demographics, and MTB strains to strengthen the
models' generalizability.
2� Proof of Concept Development: Evaluate different AI architectures specifically designed
for medical image analysis. Optimize hyperparameters and training procedures to
maximize the model's accuracy and efficiency. Employ eXplainable AI (XAI) techniques
like LIME (Local Interpretable Model-agnostic Explanations) to understand the model's
reasoning behind its diagnoses. This fosters trust and transparency among healthcare
professionals.
3� Creating Use Case Variations : Expand the AI model's capabilities to not only identify
MTB but also predict its susceptibility to different anti-tuberculosis drugs. This allows for
early tailoring of treatment regimens for optimal effectiveness. Explore the potential of
the AI model to identify other respiratory pathogens present within the same microscopic
images. This could enhance the overall diagnostic capabilities of the tool.

Develop an AI assistant that analyzes ZN-stained sputum smear microscopy images in real-time,
highlighting potential MTB bacilli to aid less experienced technicians, this is to enhance the
current alternative to culturing the mycobacterium.

22�3� Use case requirements

• AIRTBD-UC01-REQ-001: It must accurately analyze microscopic pictures of MTB-stained
sputum samples in order to detect TB infection quickly and reliably.
• AIRTBD-UC01-REQ-002: It must demonstrate high sensitivity and specificity in
differentiating MTB from other acid-fast bacilli, ensuring trustworthy diagnostic results
across diverse patient groups.
• AIRTBD-UC01-REQ-003: It must be deployable in resource-limited settings, especially
developing nations, in order to effectively address the global TB burden.
• AIRTBD-UC01-REQ-004: It requires thorough validation and regulatory certification to
assure its safety, efficacy, and conformity with medical standards and regulations.

89