This news post provides a patient friendly summary of an article published in Annals of Hematology, 25 August 2026.
Titled, ‘Deep learning differentiates myeloproliferative neoplasms and predicts JAK2 and CALR mutations from bone marrow smears’ it explores the potential of artificial intelligence (AI) to support pathologists and clinicians in their diagnoses.
What is this research about?
Researchers investigated whether AI can help clinicians and pathologists diagnose and distinguish between the three main types of MPNs: Polycythaemia vera (PV), Essential thrombocythaemia (ET) and Myelofibrosis (MF).
Specifically they examined whether AI could examine digital images of bone marrow samples and recognise patterns that help identify which type of MPN a person has.
And whether AI could predict the presence of certain genetic mutations, particularly JAK2 and CALR, from the appearance of bone marrow cells.
They used AI to examine digital images of bone marrow smears from 460 MPN patients (110 PV, 174 ET and 176 MF). The AI system was trained to recognise differences in appearance and arrangement of bone marrow cells. It was then tested to see how accurately it could: distinguish between PV, ET and MF and also predict whether certain genetic mutations were present.
What did they discover?
The results were promising.
1. AI could distinguish between the three MPN types. (accuracy of around 75%). It was particularly good at recognising PV, which it identified more reliably than ET or MF.
2. AI could also identify patterns associated with genetic mutations. The presence of certain mutations, including JAK2 and CALR, appeared to leave recognisable patterns in bone marrow cells. AI was able to predict these mutations with moderate accuracy, even without directly examining the person’s genetic test results.
This suggests that genetic changes may influence the appearance of bone marrow cells in ways that computers can recognise.
What does this mean for MPN patients?
AI could eventually become another useful tool for supporting clinicians in diagnosing and classifying MPNs. Potential benefits include:
- Helping distinguish between different types of MPNs.
- Providing additional information when bone marrow samples are difficult to interpret.
- Helping identify patients who may need further genetic testing.
- Supporting more consistent diagnoses.
- Potentially providing additional information alongside traditional bone marrow examinations, blood tests and genetic testing.
What are the limitations?
This was early-stage research, and there are several important limitations.
- The study involved only 460 patients.
- The AI was tested using images from the study rather than being independently validated in different hospitals and patient populations.
- It was less accurate at distinguishing ET from MF.
- It could not reliably predict every genetic mutation.
- It has not yet been established as a tool for routine clinical diagnosis.
The researchers emphasise that further testing and validation are needed before this technology could be used routinely in clinical practice.
The bottom line
This study demonstrates that AI may be able to recognise subtle patterns in bone marrow cells that are not always easy to distinguish through conventional examination.
It suggests that genetic mutations such as JAK2 and CALR can influence the physical appearance of bone marrow cells in ways that AI can detect.
AI is unlikely to replace bone marrow examinations, molecular testing or the expertise of haematologists and pathologists. However, it could eventually provide an additional diagnostic tool, helping doctors make more accurate and consistent MPN diagnoses.
The article ‘Deep learning differentiates myeloproliferative neoplasms and predicts JAK2 and CALR mutations from bone marrow smears’ is freely available to read in full.
