Page 103 - AI for Good-Innovate for Impact Final Report 2024
P. 103
AI for Good-Innovate for Impact
Use case – 22: AI-Rapid TB Diagnosis
Country: Tanzania 22-Medics
Organization: Sengerema Health Institute
Contact person: Gideon Ngerageza, nadsoncarliva@ gmail .com
Ndabuye Gideon, sndabuye@ gmail .com
22�1� Use case summary table
Domain Health
Problem to be Enhanced early TB diagnosis using AI from microscopic images of
addressed sputum samples, significantly reduces diagnostic turnaround time and
the start of treatment.
Key aspects of the Image segmentation, Image classification, Microscopic images of
solution sputum samples
Technology Image segmentation, image classification.
keywords
Data availability Public data: Available here for piloting
Private data: requires requests to hospitals
Metadata (type of Images – microscopic images (currently)
data) In future – Clinical data (fever, sweating, weight trend, etc ) for multi-
modality
Model Training and Training – U-Net or Mask R-CNN for segmentation, these have proven
fine tuning to be best architectures for segmentation, accompanied by image
augmentation
Fine tuning – Segment Anything Model (SAM) can be finetuned to
build a good start.
Testbeds or pilot Automated detection of tuberculosis in Ziehl-Neelsen-stained sputum
deployments smears using two one-class classifiers Link
A sputum smear microscopy image database for automatic bacilli
detection in conventional microscopy. Database for around 1320
images. Link
22�2� Use case description
22�2�1 Description
Tuberculosis (TB) remains a significant global health threat, with millions of cases reported
annually [1]. Diagnosing active TB relies heavily on culturing Mycobacterium tuberculosis
87
