Artificial intelligence could change the way we diagnose haematological malignancies in as little as 5 years, an expert has told congress.

And while this will certainly be disruptive, it shouldn’t strike fear into the hearts of haematologists, says Professor Torsten Haferlach who, quoting from leading US physician Eric Topol’s book Deep Medicine, said: ‘AI will not replace physicians, however, physicians who use AI will replace those who don’t.”

In his talk ‘from phenotype to genotype: Why whole genome sequencing and artificial intelligence are disruptive’ Professor Haferlach, who is head of the Cytomorphology Department MLL Munich Leukemia Laboratory, said that current diagnostics were based on cytomorphology, cytochemistry, histology, histoimmunochemistry, immunophenotyping, cytogenetics, FISH and an increasing number of molecular assays.

But the landscape was changing, making the need to move away from phenotyping more pressing. For example, a study by Dr David Grimwade, from the King’s College London School of Medicine, and colleagues published in Blood in 2016 illustrated the genetic complexity of AML. 

And research from American haematologist Tim Ley from Washington University School of Medicine, St. Louis, published in the NEJM found that a complex interplay of genetic events contributed to AML pathogenesis in individual patients.

“But not only do we have this landscape of disease based on the molecular and genetic background, but we need [more information] because we have more and more targeted drugs,” Prof Haferlach added. 

“We may ask what does this all have to do with artificial intelligence?…but it’s something that we use every day, when we use google maps for example… it consists of machine learning, deep learning, neural networks and pattern recognition.” 

It’s clear that Prof. Haferlach’s team at MLL is leading the way in bringing AI closer to the lab –  investigating more than 30,000 bone marrow samples per year via AI and machine learning in conjunction with regular testing to check for reliability. 

The lab currently ’employs’ 32 robots to 100 employees, compared to Germany’s 3 per 100 and 2 per 100 employees in the US.

According to Prof Haferlach, the implementation of scalable instruments, reproducible workflows and especially the implementation of artificial intelligence for readout and interpretation of data is certainly possible.

But the next step will involve the implementation of new workflows, software, cloud computing and artificial intelligence into clinical routine. 

“This can only be feasible and reproducible when today’s gold standards are tested in parallel to the new options and workflows,” he notes.

“It will take another five years to do this and during this time, the complete scenario of diagnosis in haematology will change. This for sure can be called disruptive”.