AI can detect blood cancer with high reliability
A research team from Germany has proven that artificial intelligence (AI) can detect one of the most common forms of blood cancer - acute myeloid leukemia (AML) - with high reliability.
Acute myeloid leukemia (AML) is, according to the "Competence Network Leukemia" with 3.5 new diagnoses per 100,000 inhabitants annually, "a rare disease, but the most common form of acute leukemia in Germany." A research team has now demonstrated that artificial intelligence (AI ) AML can recognize with high reliability.
One of the most common forms of blood cancer
As the German Center for Neurodegenerative Diseases (DZNE) and the University of Bonn write in a joint press release, artificial intelligence can reliably detect one of the most common forms of blood cancer - acute myeloid leukemia (AML).
Researchers at the DZNE and the University of Bonn have now proven this in a feasibility study. According to the information, their approach is based on the analysis of the gene activity of cells that are found in the blood.
Used in practice, this procedure could support conventional diagnostic methods and possibly accelerate the start of therapy. The research results were published in the "iScience" journal.
Artificial intelligence in medicine
Artificial intelligence is a much discussed topic in medicine, especially in the field of diagnostics. "We wanted to use a concrete example to examine the potential," explains Prof. Joachim Schultze, research group leader at the DZNE and head of the genomics and immunoregulation department at the LIMES Institute at the University of Bonn.
“Because this requires large amounts of data, we evaluated data on the gene activity of cells from the blood. There are numerous studies on this and the associated results are accessible via databases. So there is an enormous data pool. We have practically collected everything that is currently available, ”says the scientist.
A kind of fingerprint of gene activity
Schultze and colleagues were concerned with the “transcriptome”: a kind of fingerprint of gene activity. Because, depending on their condition, only certain genes are “switched on” in every body cell, which is reflected in the profile of gene activity. In the current study, exactly such data - it came from cells from blood samples and comprised thousands of genes - was examined.
“The transcriptome contains important information about the condition of cells. However, classic diagnostics are based on other data. We therefore wanted to find out what an analysis of the transcriptome can do with the help of artificial intelligence, that is, with algorithms that can be learned, ”explains Schultze, who is a member of the Bonn cluster of excellence“ ImmunoSensation ”.
"In the long term, we would like to apply this approach to other questions, especially in the area of dementia."
Largest data set to date for a meta study on AML
In the current study, however, the focus was on the AML. This form of leukemia leads to death within weeks without adequate treatment.
Therefore, according to the "Competence Network Leukemia", it is extremely important that therapy is started immediately after diagnosis. “The most important part of the treatment is chemotherapy with accompanying therapy to treat the side effects. Bone marrow transplantation can also occur in individual cases, ”explain the experts. Radiation therapy plays a subordinate role at AML.
The disease goes hand in hand with the increase in pathologically altered bone marrow cells, which can ultimately get into the blood. There are then healthy cells and tumor cells whose genes each have typical activity patterns.
As stated in the DZNE and the University of Bonn, all of these activity profiles were included in the analysis. Measured data from over 12,000 blood samples - these came from 105 different studies - were taken into account: the largest data set to date for a meta study on AML.
Around 4,100 of these blood samples came from people diagnosed with AML, the rest from people with other diseases or from people who had been classified as healthy.
Enormously high hit rate
The scientists fed their algorithms with parts of this data set. The input included which samples came from AML patients and which did not.
“The algorithms then looked for patterns typical of the disease in the transcriptome. This is a process that was largely automated. We speak of machine learning, ”explains Schultze. With the pattern recognition acquired in this way, further data were then analyzed and classified by the algorithms, i.e. divided into samples with AML and without AML.
“We were of course familiar with the assignment as it was recorded in the original data, but not with the software. In this respect, we were able to check the hit rate. In some processes, this was above 99 percent. We have tested various procedures from the repertoire of artificial intelligence. There was actually an algorithm that was particularly good, but the others were just behind it. ”
No practical test has yet been developed
In practice, according to the Bonn scientist, this method could support conventional diagnostic methods and help to save costs. “In principle, a blood sample that the family doctor can take and send to a laboratory for analysis could be sufficient. I would estimate that the costs are less than 50 euros. ”
The classic AML diagnostics is very extensive. According to the information, individual processes would cost a few hundred euros per run. "However, we have not yet developed a practical test. We have only shown that the process works in principle. So the foundations are laid that you can develop a test. "
Schultze emphasized that the diagnosis of AML will continue to require specialist specialists in the future. “It is about providing the experts with a tool that supports them in the diagnosis. In addition, many patients have a true odyssey before they finally land at the specialist and receive a final diagnosis there. "
Because in the early stages, the symptoms of AML can be similar to those of a severe cold. This blood cancer is a life-threatening disease that should be treated as soon as possible.
“With a blood test, as it seems possible based on our study, it would be conceivable that the family doctor already clarifies a suspected AML. And if it hardens, refer it to a specialist. The diagnosis would then possibly be made earlier than before and therapy could start earlier. ”(Ad)
Author and source information
This text corresponds to the specifications of the medical literature, medical guidelines and current studies and has been checked by medical doctors.