© Getty Images/iStockphoto/haydenbird/iStockphoto/futurezone
Tumors in the head can now be identified using the Magnetic resonance imaging (MRI) recognize well. However, it is difficult to classify them in this way. In order to be able to correctly classify the type of tumor, it is usually necessary to take a tissue sample (biopsy). The assessment of the sample by the doctor takes time and years of experience. However, the process is vital for patients. Only when the tumor has been classified can the best possible treatment be selected and initiated.
In the future, doctors could be supported in this task by adaptive algorithms and the classification based on MRI data could be made more precise, as a current study by the Karl Landsteiner Private University for Health Sciences in Krems has revealed.
An international research team around Andreas Stadlbauer from the Central Institute for Medical Radiological Diagnostics at the University Hospital St. Pölten has a artificial intelligence (AI) with MRI data from almost 170 former brain tumor patients. Using machine learning, the 5 most common brain tumors were examined. Concretely combined image data of the so-called physiological and advanced MRI, which can provide a detailed insight into the structure and metabolism of a tumor. With advanced MRI, for example, the blood flow through tissues or the distribution of water molecules in tissues is shown.
“The physiological MRI goes beyond that and shows a higher sensitivity to smaller vessels,” says Stadlbauer of the futurezone. Parameters such as vessel diameter or density as well as information on the supply of oxygen to the tumor tissue are recorded here. The combination of MRI methods enables better classification of tumors – but this results in an increased amount of data, which, according to Stadlbauer, can often only be managed by an AI.
More precise results
In the test phase, the trained artificial intelligence was fed with the MRI data of 20 current brain tumor patients and the results of the classification were compared with those of doctors. It turned out that the AI with regard to the accuracy or misclassification performs better than the trained specialists.
By contrast, human intelligence is related to the artificial specificity and sensitivity superior to the assessment. One speaks of high specificity when healthy people are reliably judged to be healthy – on the other hand, there is high sensitivity when sick individuals are reliably identified as sick.
In addition to the classification of brain tumors, the AI could also be of interest in the context of follow-up controls, says Stadlbauer. According to him, after an operation and radiation or chemotherapy, there is a follow-up with time series from image data every 3 to 6 months. Current data are compared with those of the last examination. This has the purpose, among other things, possible recurrences to recognize. This refers to tumors that reappear in the same place after successful treatment.
If there is a suspicion of a recurrence, it can usually take several months before this is confirmed. According to Stadlbauer, a recurrence can grow so large during this time that it is no longer operable. With the help of adaptive algorithms, recurrences could be detected earlier. “If the system can predict a recurrence based on the physiological image data and calculate what a tumor will look like in 3 months, life-prolonging measures could be made possible for the patient,” says the researcher.
Stadlbauer is currently working on the fact that the physiological MRT data is not processed manually as before, but automatically thanks to a neural network. Only then will the further future of such an AI become apparent. According to Stadlbauer, whether such a system for classifying brain tumors will ever establish itself depends primarily on it acceptance from radiologists. “Basically, the interest of clinics is great, but radiologists are currently still concerned that AI could take over their work,” he says.
However, the system should serve as a second opinion and support the specialist staff in their assessment – but not replace it. In general, according to Stadlbauer, such a system in medicine is a clinically useful additional application that could ensure longer survival for the patient.
one on machine learning based algorithm called icarus can distinguish between healthy and cancerous cells. ikarus was created by a research team around Altuna Akalinhead of the technology platform “Bioinformatics and Omics Data Science” at the Max Delbrück Center for Molecular Medicine in the Helmholtz Association (MDC).
The adaptive algorithm has identified a typical pattern in tumor cells that indicates cancer. Specifically, it is an unmistakable combination of genes.
ikarus has also discovered gene types that have not been unequivocally associated with cancer. In the first step of the research project, ikarus was fed and trained with data from lung and colon cancer cells – other types of tumors were added later.
First of all, ikarus’ task was to find characteristic genes in order to be able to classify the cells correctly. After his training phase, he was able to reliably distinguish between healthy and cancerous cells.
However, it is not yet possible to predict whether the method will work for all types of tumors. For the time being, artificial intelligence is also to be tested on other types of cancer.
#Artificial #intelligence #classify #brain #tumors #doctors