Artificial intelligence squares off with human intelligence.
Artificial intelligence or AI is the ability of a computer program or a machine to think and act like humans. A typical AI needs a base of knowledge to assess its environment and takes action to achieve success (1). Interestingly, it has paved its way into many areas of our everyday life already. For instance, these days in Google mails, we have the AI enabled Smart Compose option that inherently predicts our email text and response. Even our Gmail inbox uses AI powered algorithms to make sure our emails are categorized and do not end up in spam.
It turns out AI is also useful for medical diagnosis of lethal diseases. Computers can be trained to identify patterns in disease that cannot be easily identified using standard diagnostic methods. Recently, AI enabled computer algorithms have been used to discover crucial information and patterns in brain tumors. These algorithms are aimed at diagnosing tumors more accurately so that mortality rates can be reduced. Here we discuss some of the studies that have managed to show promising results in diagnosing brain tumors with an improved accuracy and speed.
Combining DNA methylation with AI in tumor classification (2018)
There are many different types of known brain tumors and this makes it challenging to standardize diagnostic methods. In an attempt to solve this issue, a team of Cancer researchers in Heidelberg managed to train an AI that can identify different types of tumors accurately (2). They used DNA methylation (a type of epigenetic alteration) as the criteria for AI to identify different types of central nervous system tumors. DNA methylation indicates the cell of origin of a tumor and can also identify the physical changes that a tumor generates.
In this study, a sample size of 2801 cancer patients was established as the starting point. The authors utilized 91 DNA methylation classes to categories these samples, using an unclassified AI algorithm. The method was then tested with over 1100 diagnostic samples. In most cases, the AI matched the diagnosis made by the standard, manual histopathological test in most cases. Interestingly, in the 12% of the cases where it didn’t, it turned out that the histopathological results obtained by the technicians were a case of misdiagnosis, and the AI based diagnosis was accurate.
The researchers have also designed a free web platform to classify the tumors online. Here the technicians can upload DNA methylation data of the samples and classify the tumors to complete their diagnosis.
Identifying signs of brain tumor with AI based blood test (2019)
Brain tumors, particularly the malignant ones like glioblastomas are incurable and have very high mortality rates worldwide. Usually the symptoms associated with brain tumors are ambiguous, and include headaches, memory loss and dizziness. The only reliable method of diagnosis is a brain scan, but the time that it takes to receive a reliable diagnosis often means that it is too late for the patient. With this new, AI-based blood test developed by a team of researchers in Edinburgh and Strathclyde, diagnosis can be done at a much earlier stage (3).
The new test uses infrared spectroscopy to analyze chemicals in the blood of a patient. AI is then used to predict the likelihood of a tumor. Out of 400 patients tested, 40 people were identified to be positive. The test also successfully identified 82% of tumors, with a low probability of false positives. In case of glioblastoma patients, it showed 92% accuracy.
Diagnosing brain tumor intraoperatively with AI (2020)
Stimulated Raman Histology (SRH) is essentially an optical imaging method that provides rapid, label-free, high resolution images of tumor tissue samples. This technique utilities internal vibrations of biomolecules to create image contrasts so that microscopic architectures of tissues are visible. This provides a superior visualization technique compared to the standard microscopic staining methods. In this study, SRH was combined with AI to expand access to expert-level intraoperative diagnosis in the ten most commonly encountered brain tumors.
In order to build the AI based tool, researchers at the University of Michigan trained the artificial neural network with more than 2.5 million samples from 415 patients (4). This was used to classify tissue into 13 categories which represent the most common brain tumors, including malignant glioblastoma, lymphoma, metastatic tumors, and meningioma.
In order to validate this neural network, the researchers sampled 278 brain tumor patients undergoing surgery at three different university medical centers in the prospective clinical trial. The samples were divided into two categories: one where diagnosis was made by a pathologist looking at histological images, and the other which used the neural network created from the image collection. The accuracy of the former was 93.9%, while the neural network provided a diagnosis with an accuracy of 94.6%.
This is remarkable, as during surgery doctors are only able to treat what they can see. With this new method, they can visualize what would otherwise be invisible. Misdiagnoses can also be largely reduced and this can improve surgical accuracy. Moreover, this AI method does not destroy the sample, so the tissue can be used again for more testing. For example, a tumor’s underlying molecular changes can improve the accuracy of diagnosis even further. Similar approaches are also likely to work in supplying timely information to surgeons operating on patients with other cancers too, including cancers of the skin, lungs and breast.
These studies look promising in treating the brain tumor patients. With a robust and accurate diagnosis, the prognosis can be assessed better and treatments can be developed. This could potentially mean that the tumors can be treated on time and can curb the fatality rates.
Antara is a computational biology researcher at the University of Groningen. She also freelances as a science writer for Science LinX,that engages the public in the research conducted by the University. Prior to that, she studied biomedical science and bioinformatics from New Delhi. Outside of work, she enjoys organising scientific and cultural events, singing and is a travelling enthusiast.