Blood cells originate from stem cell progenitors in the bone marrow, differentiating into an array of specialized cell types. In the case of the genetic disease myelodysplastic syndrome, or MDS, these blood cells don't form properly, causing uncontrolled bleeding, frequent infections, and difficulty breathing. Additionally, MDS spikes the risk of developing leukemia.
Early warning signs of MDS are often tricky to spot. Many patients present with either no or very subtle symptoms. The condition stems from disruptions to the natural development of blood cells. A closer look at blood samples from MDS patients reveals many more immature or defective blood cells than healthy, functional ones. Over time, this culminates in physiological issues such as anemia, leukopenia, and inefficient blood clotting.
Confirming an MDS diagnosis, however, remains a significant clinical challenge. Physicians typically take samples of bone marrow cells and perform a suite of genetic tests to gain insights. Now, an innovative platform leveraging the power of machine learning could see a new dawn in MDS diagnostic technologies.
Researchers at the University of Helsinki first obtained samples of bone marrow tissues from MDS patients. The cells in these tissue layers were labeled and photographed microscopically. The resulting digital images were then fed into a computational model to "teach" the algorithm how to pick up diagnostic clues that point to MDS.
The study featured in the peer-reviewed journal Blood Cancer Discovery describes how the machine learning system identifies aberrant cells in the sample. The presence of a large proportion of unusual-looking cells points towards a higher likelihood of MDS.
The concept of using computers to support diagnostic decisions is not new. However, in the real world, technology has struggled to make sense of highly complex biological data, such as images of human tissue. The researchers have overcome this hurdle, developing a method for drawing conclusions about bone marrow cells within the landscape of other cells and tissue structures.
"The study confirms that computational analysis helps to identify features that elude the human eye," said Satu Mustjoki, lead researcher on the study. "Moreover, data analysis helps to collect quantitative data on cellular changes and their relevance to the patient's prognosis."
Mustjoki and colleagues see a lot of advantages to their new platform: it's faster, more accurate, and has a much higher throughput than conventional methods. These steps forward also fit into a broader focus of the industry to completely digitize medical science. Ultimately, this translates to earlier interventions and better patient outcomes for those with MDS.