Ada, a curious seven-year-old, looks outside the window. It is a relatively humid day in Mansa, a city in northern Zambia, but it appears as though it isn’t going to rain for a couple of hours. Unaware of the threat lurking around her, Ada happily leaves her house excited to play. Just as she reaches her friends, she suddenly slaps her right arm on reflex as the female anopheles mosquito bites her. A parasite that had taken temporary refuge inside the mosquito enters Ada’s body.

Like a pirate on his ship sailing in the ocean, the parasite travels inside Ada’s bloodstream with an explicit destination port in mind- the liver. Once there, the parasite grows and matures. Now more powerful than ever, the parasite leaves the liver and enters the bloodstream again on a new voyage toward chaos and destruction. Sailing through Ada’s blood, the parasite finds and infects red blood cells. Once inside them, it rapidly divides and generates new and abundant progeny. Like air filling a balloon, the growing number of parasites inside the helpless red blood cells eventually bursts them open, freeing the invaders back into the bloodstream. The newly emerged parasites continue infecting more red blood cells, starting an endless cycle of infection and obliteration. Ada’s red blood cells are slaughtered so rapidly that she will soon enter an extreme stage of anaemia. With a size of about one-millionth of a meter, this invasive and unsparing parasite, Plasmodium Falciparum, ironically takes down its gigantic human host, causing malaria.

Malaria is a life-threatening disease that has taunted humanity with its poignant grip and ineradicable nature since the prehistoric era. Patients, if left untreated, are plunged towards kidney failure, seizures, coma, and ultimately, death. In 2020 alone, malaria caused an estimated 241 million clinical episodes worldwide and 627,000 deaths. An estimated 95% of the deaths in 2020, mostly children, were in sub-Saharan Africa, where poor people living in rural areas that lack access to health care are at a greater risk of contracting and dying from the disease.

For any disease, lack of diagnosis precludes treatment. In the case of malaria, the most definitive diagnosis  is done by examining a blood smear from the patient on a microscopic slide, a thin flat rectangular piece of glass. The smear is stained with a special stain called Giemsa, named after its inventor, the German chemist and bacteriologist Gustav Giemsa, that colours the parasite red and the cells blue. The stained slide is then visualized under a microscope, where Plasmodium Falciparum appears as a tiny pink thread floating inside the blue round red blood cells.

A rapid and accurate malaria diagnosis is crucial in preventing severe disease and mortality. It is also critical for the rational use of expensive malaria drugs in resource-poor endemic regions. But despite available standard diagnostic methods, diagnosing malaria remains challenging due to the lack of pathologists and microscopes in rural areas. Africa has approximately one pathologist per million people, and although there is a worldwide shortage of them, many countries in the continent are hit more severely than the rest of the world. For example, on average, there are seven pathologists per 10 million people in Zimbabwe and Zambia. The situation is even worse in Mozambique, where the ratio is four pathologists for 10 million people. In comparison, the United States has an average of four pathologists for a hundred thousand people. Intrigued and motivated by this challenge, researchers from the University of Helsinki and Karolinska Institutet came together to develop a digital pathology and AI (Artificial intelligence)-based malaria diagnostics as a solution.

AI is the ability of computers to perform any task that typically requires human intelligence. Most people are familiar with Google Maps and chatbots based on AI algorithms. However, AI has much broader applications, including diagnosing malaria. The researchers’ proposed solution would start with field workers collecting blood smears from malaria patients in endemic regions. These slides would then be stained and scanned, and the image digitalized and uploaded to a cloud-based storage space. The AI algorithm would then access these images, analyzing and ranking each red blood cell, often more than 50000, according to the probability of infection depending on whether it sees a parasite inside the cells. The AI would then create a panel of images it thinks are infected and would send it to an expert for final evaluation. With this digital AI-based system, the pathologist could be anywhere and remotely diagnose patients. This solution would also limit the hours a pathologist spends carefully examining slides, saving time and money.

But how does a computer know which red blood cells are infected with a parasite and which aren’t? Like every expert in the world, it needs to be taught first. For any algorithm to predict something, it must first be trained with practice sets. The scientist first collects an extensive set of sample images and tells the algorithm which are parasite-positive and which aren’t. Then, the algorithm is given a collection of images for prediction. Like in an exam, it is asked to go through these images and make a prediction based on the training. Once the algorithm makes one, the scientist checks to ensure it is accurate. If the prediction is incorrect, better-quality training sets are used to train the algorithm, which is then tested again. This cycle is repeated until the predictions are satisfactory for the algorithm to be used in a clinical setting. With better AI systems, even automated diagnosis could be possible in the future.

In the initial pilot study in 2014, the researchers from Finland and Sweden stained 27 plasmodium-infected and 20 uninfected blood smear samples from individuals from Helsinki University Central Hospital, Finland, and digitally scanned the images. They first trained the algorithm on digital slides from ten patients and validated the quality of the training by testing it on six samples. The algorithm was then asked to analyze 19 infected and 12 non-infected samples and make an accurate diagnosis. In parallel, a pathologist was asked to do the same to compare the results. The AI achieved a diagnostic specificity of 100%, correctly identifying all the red blood cells infected by the parasite.

Banking on the initial success of the pilot study, in 2019 the researchers tested their remote digital pathology and AI algorithm on blood smears from 125 malaria patients in rural Tanzania, a country in East Africa. Though the results were very promising, the researchers observed a significant constraint in implementing this system in remote rural, malaria-endemic settings. Currently available scanners for digitalizing good-quality images are big, bulky, and expensive. Very few portable mini scanners are commercially available, but even these are limited by their cost and low-image resolution. To address this challenge, the Finnish research team is now developing and testing low-cost, high-resolution mini scanners. Perhaps one day, such scanners will be seen worldwide as part of routine point-of-care devices.

Over the years, the Helsinki research team has also been working towards developing AI-based tools for cervical cancer screening and breast and thyroid cancer detection-based applications. The technology has also been successfully implemented by other teams in radiology, oncology, and surgical pathology. With rapid development in highly advanced computer processing power and software technology, the implications of AI in medicine are expected to grow and impact our future lives radically. Including many like the seven-year-old Ada.

REFERENCES

1) Milner, D. A. (2018). Malaria pathogenesis. Cold Spring Harbor perspectives in medicine, 8(1), a025569.

2) Global Malaria Programme (2021, December 6). World malaria report 2021. https://www.who.int/publications/i/item/9789240040496

3) Mathison, B. A., & Pritt, B. S. (2017). Update on malaria diagnostics and test utilization. Journal of clinical microbiology, 55(7), 2009-2017.

4) Dafaallah, K., & Awadelkarim, A. A. (2010). Role of pathology in sub-Saharan Africa: an example from Sudan. Pathol Lab Med Int, 2, 49-57.

5) Mudenda, V., Malyangu, E., Sayed, S., & Fleming, K. (2020). Addressing the shortage of pathologists in Africa: Creation of a MMed Programme in Pathology in Zambia. African Journal of Laboratory Medicine, 9(1), 1-7.

6) Ahmad, Z., Rahim, S., Zubair, M., & Abdul-Ghafar, J. (2021). Artificial intelligence (AI) in medicine, current applications and future role with special emphasis on its potential and promise in pathology: present and future impact, obstacles including costs and acceptance among pathologists, practical and philosophical considerations. A comprehensive review. Diagnostic Pathology, 16(1), 1-16.

7) Linder, N., Turkki, R., Walliander, M., Mårtensson, A., Diwan, V., Rahtu, E., … & Lundin, J. (2014). A malaria diagnostic tool based on computer vision screening and visualization of Plasmodium falciparum candidate areas in digitized blood smears. PLoS One, 9(8), e104855.

8) Holmström, O., Stenman, S., Suutala, A., Moilanen, H., Kücükel, H., Ngasala, B., … & Lundin, J. (2020). A novel deep learning-based point-of-care diagnostic method for detecting Plasmodium falciparum with fluorescence digital microscopy. Plos one, 15(11), e0242355.

9) Holmström, O., Linder, N., Moilanen, H., Suutala, A., Nordling, S., Ståhls, A., … & Lundin, J. (2019). Detection of breast cancer lymph node metastases in frozen sections with a point-of-care low-cost microscope scanner. Plos one, 14(3), e0208366.

10) Holmström, O., Linder, N., Kaingu, H., Mbuuko, N., Mbete, J., Kinyua, F., … & Lundin, J. (2021). Point-of-care digital cytology with artificial intelligence for cervical cancer screening in a resource-limited setting. JAMA network open, 4(3), e211740-e211740.

11) Stenman, S., Linder, N., Lundin, M., Haglund, C., Arola, J., & Lundin, J. (2022). A deep learning–based algorithm for tall cell detection in papillary thyroid carcinoma. PloS one, 17(8), e0272696.