AI-based Detection of Malaria: Overview and Challenges
FGAI4H-H-014-A03
Brasilia, 22-24 January 2020
2
Microscopy Diagnosis
Gold standard for diagnosis of malaria is a microscope
3
Malaria burden
Currently in Uganda, and many
developing countries
•
Malaria has been reported as one of
the leading cause of death accounting
for over 27\% of lives of Ugandans
•
Patient in big numbers wait to be
diagnosed
4
Microscopy challenge
In highly malaria endemic countries;
•
There is lack of enough trained lab
technicians
•
In Ghana, 1.72 microscopes per
100,000 population, but only 0.85
trained laboratory staff per 100,000
population
5
Microscopy challenge
•
Standard Operating Procedure
requires not to view more than 20
slides a day
•
Microscopy is eye straining
•
Less diagnosis throughput
•
Variations in individual expert
judgment
6
Use case detail
7
Image capture detail
8
Image capture
9
Captured image
•
1182 images captured by smart phone and 2703 captured by Motic camera.
•
The blue bounding box drawn around a pathogen
Thick blood smear microscopic image captured with a smartphone
10
Captured image
Image Annotation by experts-Binary classification task
Expert annotation process
11
Binary classification
Malaria Positive patches are annotated by drawing bounding boxes around the pathogens
in
(1182 images captured by smart phone and 2703 captured by motic camera)
Positive patches
labelled as "malaria"--TP
Negative patches:
anything in the image not labelled as malaria--TN
Used
PASCAL VOC format
here for annotation as a benchmark
12
AI APPROACH-Supervised Machine Learning
Approaches for benchmark;
Traditional machine learning(with feature engineering)-Extra Random Tree
Deep learning approach-Convolutional Neural Network
13
Bench mark deep learning architecture-CNN
AI approach using CNN architecture
https://www.sciencedirect.com/topics/engineering/convolutional-neural-networks
14
Positive and negative pathogen CNN accuracies
Patch detection accuracies
Plasmodium patch detection accuracies
15
Deep learning accuracies (CNN) VS traditional machine learning (Extra Random Tree(ERT))
CNN VS ERT ROC accuracies for malaria(a), TB(b) and Hookworm
ROC AUC for malaria, TB and intestinal parasites detection
16
CNN Algorithm performance (red boxes) Vs expert annotation(white)
Detected Objects for plasmodium (left) and bacilli(right).
17
User Interface
Web based testing prototype
18
Before detection
Test image upload
19
After detection
Test image upload
20
Thank You
Rose Nakasi
g.nakasirose@gmail.com
END
This presentation delves into the utilization of AI for malaria detection, showcasing the background that led to the creation of TG-Malaria. Highlighting the current burden of malaria in Uganda and other developing nations, it addresses challenges in microscopy diagnosis due to a lack of trained technicians. The slides depict the need for innovative solutions, such as image capture and expert annotation for binary classification tasks. Emphasizing the significance of AI in enhancing diagnostic processes, the presentation sheds light on the potential to revolutionize malaria diagnosis in resource-constrained settings.
Download Presentation
Please find below an Image/Link to download the presentation.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author. Download presentation by click this link. If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.
E N D
Presentation Transcript
FGAI4H-H-014-A03 Brasilia, 22-24 January 2020 Source: TG Malaria Topic Driver Title: Overview of the topic AI-based detection of malaria Purpose: Information Contact: Rose Nakasi E-mail: g.nakasirose@gmail.com Abstract: This PPT provides background information that led to the creation of the TG-Malaria and is provided as information to the FG-AI4H participants at the Brasilia meeting.
Microscopy Diagnosis Gold standard for diagnosis of malaria is a microscope 2
Malaria burden Currently in Uganda, and many developing countries Malaria has been reported as one of the leading cause of death accounting for over 27\% of lives of Ugandans Patient in big numbers wait to be diagnosed 3
Microscopy challenge In highly malaria endemic countries; There is lack of enough trained lab technicians In Ghana, 1.72 microscopes per 100,000 population, but only 0.85 trained laboratory staff per 100,000 population 4
Microscopy challenge Standard Operating Procedure requires not to view more than 20 slides a day Microscopy is eye straining Less diagnosis throughput Variations in individual expert judgment 5
Captured image 1182 images captured by smart phone and 2703 captured by Motic camera. The blue bounding box drawn around a pathogen Thick blood smear microscopic image captured with a smartphone 9
Captured image Image Annotation by experts-Binary classification task Expert annotation process 10
Binary classification Malaria Positive patches are annotated by drawing bounding boxes around the pathogens in (1182 images captured by smart phone and 2703 captured by motic camera) Positive patches labelled as "malaria"--TP Negative patches: anything in the image not labelled as malaria--TN Used PASCAL VOC format here for annotation as a benchmark 11
AI APPROACH-Supervised Machine Learning Approaches for benchmark; Traditional machine learning(with feature engineering)-Extra Random Tree Deep learning approach-Convolutional Neural Network 12
Bench mark deep learning architecture-CNN AI approach using CNN architecture https://www.sciencedirect.com/topics/engineering/convolutional-neural-networks 13
Plasmodium patch detection accuracies Positive and negative pathogen CNN accuracies Patch detection accuracies 14
ROC AUC for malaria, TB and intestinal parasites detection Deep learning accuracies (CNN) VS traditional machine learning (Extra Random Tree(ERT)) CNN VS ERT ROC accuracies for malaria(a), TB(b) and Hookworm 15
Detected Objects for plasmodium (left) and bacilli(right). CNN Algorithm performance (red boxes) Vs expert annotation(white) 16
Web based testing prototype User Interface 17
Test image upload Before detection 18
Test image upload After detection 19
END Thank You Rose Nakasi g.nakasirose@gmail.com 20
