Deep Learning Technology Improves Diagnostic Pathway for Brain Tumors

Simplified Overview of Brain Tumors

According to Johns Hopkins Medicine, over 120 types of brain tumors, depending on the tissue affected, are known. While brain tumors do not necessarily equate to brain cancer, benign tumors in some brain regions can be dangerous, even life-threatening.

Malignant brain tumors are tumors that proliferate and spread to healthy tissue. These types of tumors are considered cancerous. Primary and metastatic brain tumors are subcategories of malignant brain tumors.

Primary brain tumors are cancerous or malignant and originate in the brain. While these tumors can spread to other parts of the brain and spinal cord, metastatic brain tumors are more common. Metastatic brain tumors originate in other body areas and spread to the brain. These are also sometimes referred to as secondary brain tumors. Breast, colon, kidney, lung, and skin cancer can commonly metastasize to the brain.

Symptoms of a brain tumor can vary drastically based on its location, size, and growth rate. Taking that into consideration, Johns Hopkins Medicine notes that the most common symptoms of brain tumors include headaches, seizures, cognitive issues, aphasia, behavioral changes, numbness or paralysis, motor issues, visual changes, and memory loss.

Diagnostic Methods

According to the article in ACS Nano, 90% of all central nervous system tumors are in the brain. Additionally, the publication notes that the five-year survival rate of brain tumors is as little as 36%. Despite the prevalence, little headway has been made in diagnosing and treating brain tumors early. This is partially due to the brain’s sensitivity and limited understanding of diagnosing and treating neurological conditions without risking additional damage and side effects.

Current methods to diagnose brain cancer rely heavily on imaging and tissue biopsy to identify the presence of a brain tumor, determine its location, and develop a treatment plan.

In the article, Premachandran and her colleagues criticize the limitations of these current diagnostic methods. They note that tissue biopsy is highly invasive, clinically challenging, difficult to repeat, and unable to capture tumor heterogeneity. While imaging may be less complex, it is accompanied by limited sensitivity and resolution concerns.

An Overview of the Publication

The paper released by Premachandran and Tan focused on a noninvasive liquid biopsy accompanied by deep surveillance of brain cancer.

“To generate comprehensive information on a brain tumor, it is necessary first to detect the presence of a malignant tumor, second, identify whether the tumor has a primary or secondary origin, and third, to find the exact part of the brain where the tumor is housed. Such deep analysis is essential for meaningful and timely precision medicine,” stated Premachandran, Tan, and their peers in the publication.

Current Diagnostic Methods and Biomarkers

Premachandran notes that the brain tumor location is typically determined through imaging and followed up with a tissue biopsy — usually a lengthy, painful, and risky process. Tissue biopsy also has limitations as it cannot be performed at several brain locations because of its complexity.

Diagnostic blood tests and bioassays to detect biomarkers have been used for other kinds of cancer. In the publication, Premachandran writes that biomarkers such as ctDNA, microRNA, tumor-associated platelets, and extracellular vessels have already been explored for cancer diagnosis.

“They have focused on two types of biomarkers: cell-free DNA and DNA methylation,” noted Tan. “The thing with this approach is they need quite a large quantity of blood.”

She also shared that there is no published evidence on the efficacy of bioassays in detecting brain cancers. “They can do a more common type of cancer like breast or lung cancer. But nobody has reported any attempt to do brain cancer. Brain cancer is tough to detect because of the brain–blood barrier,” added Tan. “The biomarker has a very low concentration in blood.”

The Nanosensors

The unique diagnostic approach proposed by Tan, Premachandran, and the rest of the researchers involves using nanosensors developed by the team. Tan and her colleagues fabricated the nanomaterials and were looking for ways to apply them. Eventually, they found that the most promising application for these nanomaterials was for noninvasive liquid brain biopsies.

“My team and I use a different approach. We don't use bioassays; we use nanomicroscopy coupled with the nanosensor we generated in our lab,” Tan said.

She shared that there are three components to the technology. The first — and arguably the most important — component is the nanosensor. Tan and Premachandran emphasized that their sensors are highly sensitive compared to current ongoing research.

“Our contribution here is that we have very sensitive sensors. They can detect all the previously undetectable biomarkers,” added Tan.

Tan also outlined that, while many nanosensors are on the market, many are unstable, making them unrealistic for practical use. “The sensors we test in our lab have good repeatability and reproducibility. We think that, from an engineering point of view, we can make a reliable, stable instrument out of it,” Tan expanded. “Making this technique more readily commercialized starts with developing a stable and sensitive biosensor.”

“The second part of this technique is developing new biomarkers,” continued Tan. “Because we have very sensitive sensors, we can find new biomarkers, which has never been attempted before.”

Deep Learning

“The third part is the machine learning computerized diagnosis. This is the deep surveillance part of the paper,” said Tan. “We have multiple biomarkers and can detect them with one test. So how do healthcare professionals use these biomarkers for diagnosis?”

Tan emphasized that deep learning technology does more than provide patients with information on whether they have cancer. In this case, deep learning technology may even provide clinicians with the location of brain cancer.

Premachandran told LifeSciencesIntelligence that the neural learning network is what can help predict tumor location. Premachandran explained that machine learning is a computerized process. She notes, “deep learning means allowing the computer to study what it is given and then predict the outcome given a real-life sample.”

“The research team trains the model and then lets it recognize the pattern. Then the model can predict the tumor’s location,” explained Premachandran. “It would be very advantageous because we are doing it from a minimal amount of blood.”

According to the publication, Premachandran and her collaborators found that detecting primary and secondary tumors by the deep learning model was 100% accurate. Additionally, the article notes that the model predicted intracranial tumor locations with 96% accuracy.

Benefits of the Approach

The most apparent benefit of this tool is that it’s less invasive than a tissue biopsy. “The blood is taken, just like a regular blood test. It's not from a challenging access point,” noted Tan. The patient experience should be similar, if not the same, to any other blood test.

Offering a less invasive approach will ease the burden and risks associated with a tissue biopsy in the brain.

One might argue that noninvasive diagnostic methods are available with imaging; however, the limited sensitivity and resolution present a diagnostic barrier that can be overcome with this approach. As mentioned, the deep learning models yielded perfect and nearly perfect detection of tumors and prediction of location.

There is also data to support that this tool has been effective for other cancer forms. “Brain cancer detection is part of a big project. We also tested on other cancer types, like breast cancer, lung cancer, and colorectal cancer. We have published data on those cancer types already. Right now, the research is focusing on early-stage cancer diagnosis. With diagnosis earlier, we can treat earlier and have better patient outcomes,” commented Tan.