Tackling Malaria Diagnosis in sub-Saharan Africa with Fast, Accurate and Scalable Robotic Automation, Computer Vision and Machine Learning (FASt-Mal)
Lead Research Organisation:
University College London
Department Name: Computer Science
Abstract
Malaria affect about 300 million people worldwide leading to around one million deaths each year. Up to eighty-five percent of the cases occur in sub-Saharan Africa with about 90% mortality in the under five years-of-age group due to severe malaria syndromes. Control of malaria remains a major public health issue in sub-Saharan Africa developing countries. A quarter of the global malaria cases and a third of malaria-attributable childhood deaths occur in the most populous country of Africa, Nigeria (160M inhabitants) and indicates the importance of the problem. Accurate malaria diagnosis relies on the recognition of clinical parameters and more importantly in the microscopic detection of malarial parasites, parasitised red-blood-cells in peripheral-blood films. Malaria parasite detection and counting by human-operated optical microscopy is the current "gold standard" and despite its major severe drawbacks, other non-microscopic methodologies have not been able to outperform it. Presumptive treatment for malaria (without microscopic confirmation) is wasteful of drugs and ineffective if the diagnosis was wrong, a drain on often precious health resources, fuels antimalarial resistance and have made control and elimination interventions unachievable. We aim to create and test in real-world conditions a fast, accurate and scalable malaria diagnosis system by replacing human-expert optical-microscopy with a robotic automated computer-expert system FASt-MalPrototype that assesses similar digital-optical-microscopy representations of the problem. The system aims to provide access to effective malaria diagnosis, a challenge that is faced by all developing countries where malaria is endemic.
Planned Impact
The global incidence of clinical malaria is estimated at about 300 million cases leading to around one million deaths each year. Up to eighty-five percent of the cases occur in sub-Saharan Africa with about 90% mortality in the under five years-of-age group due to severe malaria syndromes. Control of malaria remains a major public health issue in sub-Saharan Africa developing countries. Our proposal addresses a key problem primarily related to these large group of low- and low-to-middle income countries with large burden Global Health Challenges such a malaria. A quarter of the global malaria cases and a third of malaria-attributable childhood deaths occur in the most populous country of Africa, Nigeria (160M inhabitants) and indicates the importance of the problem. Large all-year-round lethal malaria morbidity and mortality burden has hindered Ibadan wellbeing and economic development within the South-West Nigerian geopolitical region. Key strategies for malaria control has been to reduce mortality by rapid treatment with antimalarial drugs, but this has been stalled by the lack of scalable accurate diagnosis methods. Human-microscopic examination of blood smears remains the "gold standard" for malaria diagnosis and despite its major severe drawbacks, other non-microscopic methodologies have not been able to outperform it. Access to effective malaria diagnosis is a challenge faced by all developing countries where malaria is endemic. Presumptive treatment for malaria (without microscopic confirmation) is wasteful of drugs and ineffective if the diagnosis was wrong, a drain on often precious health resources, fuels antimalarial resistance and have made control and elimination interventions unachievable. This has prompted WHO, CDC and other Global Health organisations to emphasise the urgent need for tools to overcome the deficiencies of human-operated optical-microscopy malaria diagnosis. Our research aims to provide a novel solution for automated fast accurate scalable computational optical-microscopy identification of malaria parasites that will certainly underpin the design and development of future portable accurate and cost-effective malaria-detection devices. Our proposal has a clear path to immediate- and near-future impact outcomes across malaria endemic regions. The proposed timeline is to achieve the immediate-future impact outcome: the design and deployment of a cheaper and optimised robotic bench-top prototype at our primary beneficiary overseas partner at COMUI. Our Nigeria-UK team will carry-out a novel large-scale preclinical assessment of the validity of the automated computational system in real world clinical conditions. The COMUI University College Hospital Ibadan is a centre of academic excellence and attracts both wellbeing and ill people seeking affordable good quality healthcare. Our population footprint has allowed us to execute activities to engage users from different socio-economic backgrounds. Our engagement with primary users is extremely high given to the all-year-round burden of malaria at all ages of the population of the Ibadan metropolis. All inhabitants are at high risk of malaria infection with the most affected being pregnant women, children and the elderly. Fast accurate and scalable malaria-diagnosis methods such as ours will improve health and wealth on large sectors of the population living in extreme poverty across Nigeria and more importantly across the sub-Saharan Africa region (i.e. Togo, Ghana, Cameroon among others).The impact of our research is enormous as rapid and reliable accurate diagnosis of malaria, and therefore its accurate prompt treatment, is a crucial challenge for fulfilling the international development goals for the sub-Saharan African region and other regions of the World affected with malaria.
Publications
Marquet S
(2017)
A Functional IL22 Polymorphism (rs2227473) Is Associated with Predisposition to Childhood Cerebral Malaria.
in Scientific reports
Manescu P
(2022)
Content aware multi-focus image fusion for high-magnification blood film microscopy.
in Biomedical optics express
Brown BJ
(2020)
Data-driven malaria prevalence prediction in large densely populated urban holoendemic sub-Saharan West Africa.
in Scientific reports
Abah SE
(2020)
Depleted circulatory complement-lysis inhibitor (CLI) in childhood cerebral malaria returns to normal with convalescence.
in Malaria journal
Manescu P
(2023)
Detection of acute promyelocytic leukemia in peripheral blood and bone marrow with annotation-free deep learning.
in Scientific reports
Claveau R
(2020)
Digital refocusing and extended depth of field reconstruction in Fourier ptychographic microscopy.
in Biomedical optics express
Manescu P
(2020)
Expert-level automated malaria diagnosis on routine blood films with deep neural networks.
in American journal of hematology
Bendkowski C
(2021)
Histological and cytological imaging using Fourier ptychographic microscopy
Moysis E
(2024)
Leveraging deep learning for detecting red blood cell morphological changes in blood films from children with severe malaria anaemia.
in British journal of haematology
Lou D
(2022)
LncPheDB: a genome-wide lncRNAs regulated phenotypes database in plants.
in aBIOTECH
Abah SE
(2018)
Low plasma haptoglobin is a risk factor for life-threatening childhood severe malarial anemia and not an exclusive consequence of hemolysis.
in Scientific reports
Pérez-Ortiz M
(2022)
Network topological determinants of pathogen spread.
in Scientific reports
Shaw M
(2021)
Optical mesoscopy, machine learning, and computational microscopy enable high information content diagnostic imaging of blood films.
in The Journal of pathology
Fisher T
(2023)
Specialist hybrid models with asymmetric training for malaria prevalence prediction.
in Frontiers in public health
Claveau R
(2020)
Structure-dependent amplification for denoising and background correction in Fourier ptychographic microscopy
in Optics Express
Turiel J
(2021)
Wisdom of crowds detects COVID-19 severity ahead of officially available data.
in Scientific reports
Description | We have discovered key aspect of robotic-microscopy and machine learning related to solving the challenge of automated diagnosis of malaria. We have now developed two microscopy strategies. One based on the capture-fastmal scanning microscope and a novel one based on fourier ptycography which is likely to have a tremendous impact in deploying the system in sub-saharan Africa. The fourier ptycography microscope is fast in acquiring data, provides high-resolution of a wide area of the specimens and more importantly it has no-moving parts. The ML framework of the digital pathology system is underway with the expected difficulties of a challenging real-world problem. We have developed new method for Deep Learning Extended Depth of Field Fusion and object detectors for the diagnosis of a specimen. We are in pace in digitising this large and unique image and clinical dataset. We have now designed, implemented and validated a DeepLearning approach named Deep Malaria Convolutional Neural Network (DeepMCNN) that has PPV and NPV suitable for clinical use in the sub-Saharan malaria region. We are also in the process to deploy a smart AI driven navigational tool for the microscope. |
Exploitation Route | The findings are part of the development of an automated malaria diagnosis system that will be deployed across our health provision centres in Ibadan, Nigeria, an urban metropolis of 3-million inhabitants with large malaria burden. From there we attempt to deploy across the region. This will have a massive impact in malaria diagnosis capacity, capability and reliability. We also have plans to deploy a novel ptychographic microscope that could speed up diagnosis time. The DeepMCNN is being deployed in the clinical site in sub-Saharan West Africa as well as a simpler version based on mobile smartphone cameras and GPU. |
Sectors | Aerospace Defence and Marine Communities and Social Services/Policy Digital/Communication/Information Technologies (including Software) Electronics Financial Services and Management Consultancy Healthcare Government Democracy and Justice Manufacturing including Industrial Biotechology Pharmaceuticals and Medical Biotechnology Other |
URL | https://african-cschd.org |
Description | Our work and findings related to this award are having an positive immediate effect on all our patient cohorts at the College of Medicine of University of Ibadan. The standardisation and procedural adjustments deployed as an immediate consequence of this research have increased the capability and capacity as well as the quality of malaria diagnosis across the users of our developed technology on the urban metropolis of Ibadan Nigeria where 3-million inhabitants live under intense malaria burden all-year-round. Data from this diagnostic pipeline is allowing us to validate data-driven malaria prevalence prediction systems which we have now published. This is an ideal integration of the aim of the project (digital enabled malaria diagnosis) with rapid clinical benefit to patient with the use of such information for public and global health monitoring. We have created an African Computational Sciences for Health and Development where expertise from this and other projects with our African partner are showcased and facilitate the access to the region. As a result of our research and impact activities supported by this grant, we have contributed to the academic promotion to full Professor of computer science of an academic based at the University of Ibadan. We have now established contact with NGO focused on training of Machine Learning and Artificial Intelligence in the African Region. We have established links with African Institute of Mathematics and together with University of Ibadan (UI) Computer Science, University of Tübingen and us at UCL Computer Science we have obtained PhD funds for a Nigerian student based and registered at UI Computer Science working on AI for biomedicine and healthcare that is relevant for the sub-Saharan region. These activities embodies our commitments to equality, diversity, and inclusion and ensuring that the benefits of AI and machine learning are accessible across different geographies and demographics. Our findings and further impact promote global engagement and creating opportunities for underrepresented groups in STEM fields, thereby enriching the academic and research landscape with a broader range of voices and perspectives. |
First Year Of Impact | 2017 |
Sector | Communities and Social Services/Policy,Creative Economy,Digital/Communication/Information Technologies (including Software),Education,Healthcare,Manufacturing, including Industrial Biotechology,Pharmaceuticals and Medical Biotechnology,Other |
Impact Types | Cultural Societal Economic Policy & public services |
Description | Data driven disease burden prediction for clinical service management |
Geographic Reach | Africa |
Policy Influence Type | Influenced training of practitioners or researchers |
Impact | We sustain the management of resources via our data driven approach. |
Title | Giemsa Stained Thick Blood Films for Clinical Microscopy Malaria Diagnosis with Deep Neural Networks Dataset. |
Description | Giemsa Stained Thick Blood Films for Clinical Microscopy Malaria Diagnosis with Deep Neural Networks Dataset. |
Type Of Material | Technology assay or reagent |
Year Produced | 2020 |
Provided To Others? | Yes |
Impact | ---------------------------------------------------------------------------------- Context: Thick Blood Films (TBF) remains the gold standard for diagnosing malaria in sub-Saharan regions. TBF relies on the availability of a trained human microscopist to visually inspect Giemsa stained blood smears under a light microscope to identify and count the P. falciparum parasites. This is time-consuming and subject to human error. A wrong diagnosis of malaria can have negative consequences for patients and for anti-malarial therapy resources. Over-treatment, in the long run, leads to parasite resistance. ---------------------------------------------------------------------------------- Dataset Image acquisition: Image fields from Giemsa stained thick blood smears were captured using a upright brightfield microscope (Olympus BX63) fitted with a 100X/1.4NA objective lens, a motorised stage (Prior Scientific) and a colour camera (Edge 5.5c, PCO). Each image field covers an area of 166 µm x 142 µm (2560x2160 pixels). A z-stack comprising 14 focal planes with a separation of 0.5 µm was captured at each position with a camera exposure time of 50 ms. The z-stacks were projected onto a single plane using a wavelet-based extended depth of field algorithm. |
URL | https://rdr.ucl.ac.uk/articles/dataset/Giemsa_Stained_Thick_Blood_Films_for_Clinical_Microscopy_Mala... |
Title | Case-Control Severe Malaria Study Clinical Database |
Description | Database of large case-control study focused on severity biomarker discovery SQL database with clinical, laboratory and experimental data |
Type Of Material | Database/Collection of data |
Provided To Others? | No |
Impact | This large study is producing new findings of molecules that correlate with childhood severe malaria |
Title | FASt-Mal Malaria Parasite Object Label Expert Database |
Description | Expert microscopist object-level labels on digitised malaria thick blood films |
Type Of Material | Database/Collection of data |
Year Produced | 2020 |
Provided To Others? | Yes |
Impact | This database that will be available upon our publication due 2020 will be available to train deep learning object detectors for malaria artificial intelligence assisted devices. |
URL | https://african-cschd.org/index.php/75-FASt-Mal |
Title | High Magnification Z-Stacks from Blood Films |
Description | All biological samples were collected from participants recruited under the auspices of the CMRG at the 800-bed tertiary hospital, UCH in the city of Ibadan, Nigeria, after after de-identification of the patient information. We trained and tested the CAMI-Fusion network on three different types of samples: (a) Giemsa-stained thick blood films (TBF) and (b) Giemsa-stained peripheral blood smears (PBS) for malaria parasite detection as well as (c) Wright-stained Bone Marrow Aspirates (BMA) for diagnosing blood cancers.Image acquisition and pre-processing: We used an upright brightfield microscope (BX63, Olympus, 100W halogen bulb light source) with a 100X/1.4NA oil immersion objective (MPlanApoN, Olympus) and a color digital camera (Edge 5.5C, PCO) to acquire multiple high-resolution z-stacks with an axial step of a 0.5 µm. White balancing was applied to each focal plane image. For each sample acquisition, an empty field of view (empty region of the slide not containing any blood cells) was acquired and considered as a white object reference. |
Type Of Material | Database/Collection of data |
Year Produced | 2022 |
Provided To Others? | Yes |
URL | https://rdr.ucl.ac.uk/articles/dataset/High_Magnification_Z-Stacks_from_Blood_Films/13402301 |
Title | Malaria Prevalence in Large Densely-Populated Urban Holoendemic sub-Saharan West Africa: The Ibadan 1996 to 2017 Dataset |
Description | The Ibadan, Nigeria malaria-prevalence dataset 1996 to 2017.
When using the dataset please also cite: Brown, B.J., Manescu, P., Przybylski, A.A. et al. Data-driven malaria prevalence prediction in large densely populated urban holoendemic sub-Saharan West Africa. Sci Rep 10, 15918 (2020). https://doi.org/10.1038/s41598-020-72575-6 For metadata and supplementary information see:Brown, B.J., Manescu, P., Przybylski, A.A. et al. Data-driven malaria prevalence prediction in large densely populated urban holoendemic sub-Saharan West Africa. Sci Rep 10, 15918 (2020). https://doi.org/10.1038/s41598-020-72575-6 |
Type Of Material | Database/Collection of data |
Year Produced | 2020 |
Provided To Others? | Yes |
URL | https://rdr.ucl.ac.uk/articles/dataset/Malaria_Prevalence_in_Large_Densely-Populated_Urban_Holoendem... |
Description | African Institute for Mathematical Sciences |
Organisation | African Institute for Mathematical Sciences |
Country | South Africa |
Sector | Academic/University |
PI Contribution | We obtained funding for an African PhD student to do a PhD registered at the University of Ibadan and supervised by University of Ibadan, University of Tübingen and UCL academics on AI for Biomedicine and Healthcare. |
Collaborator Contribution | AIMS provided the funding, UI, UT and UCL provide the academic supervision and access to relevant resources at the College of Medicine of the University of Ibadan. |
Impact | Currently the PhD student based in UI, Nigeria has been working on transferring AI models to be used in improving retinal disease in relevant cohorts at University College Hospital Ibadan, Nigeria. |
Start Year | 2020 |
Description | Kabarak University Digital Pathology |
Organisation | Kabarak University |
Country | Kenya |
Sector | Academic/University |
PI Contribution | We have exchanged protocols to deploy a pan-African Digital Innovation network in Digital Pathology |
Collaborator Contribution | We have exchanged protocols to deploy a pan-African Digital Innovation network in Digital Pathology |
Impact | Work still in progress to identify what pathologies will be relevant for the digital pathology framework to be used. |
Start Year | 2019 |
Description | University of Dodoma Labelling App |
Organisation | University of Dodoma |
Country | Tanzania, United Republic of |
Sector | Academic/University |
PI Contribution | We have deployed some of fastmal algorithms and protocols with a team in East Africa University of Dodoma |
Collaborator Contribution | They have contributed with samples and annotations of malaria specimens as well as the deep learning platform |
Impact | we are still working on the labels and the east African specimens |
Start Year | 2019 |
Title | AUTOMATED MICROSCOPY |
Description | A computer implemented method of controlling a microscope (632) is provided. The method comprises capturing an image (631) within a field of view of a lens of the microscope (632) configured to view a sample on a motorised stage (633) of the microscope (632). The image comprises a portion of the sample. The image (631) is provided to an artificial neural network (610). An action (611) for moving the motorised stage (633) is determined in dependence on an output of the artificial neural network (610). The motorised stage (633) is moved automatically in accordance with the action (611). |
IP Reference | WO2022129867 |
Protection | Patent / Patent application |
Year Protection Granted | 2022 |
Licensed | Commercial In Confidence |
Impact | The proposition of a competitive method to deal with sampling bottlenecks for optical microscopy of blood films. |
Description | Nigeria Health Research Capacity Network, Abuja |
Form Of Engagement Activity | Participation in an activity, workshop or similar |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Professional Practitioners |
Results and Impact | We co-organised the workshop together with the Wellcome Trust, The Nigerian Institute for Medical Research, University of Ibadan and many other international partners such as CDC Nigeria. The working team led to the foundation of a fund to support the training of healthcare researchers to successfully bid for funding internationally. |
Year(s) Of Engagement Activity | 2018 |