Source: Artvizual/Pixabay

An area of great hope and promise for applied artificial intelligence (AI) deep learning is at the intersection of neuroscience and oncology, both challenging fields known for their inherent complexity. A new study published in Biology Methods & Protocols demonstrates how the unique combination of explainable AI (XAI) and repurposing camouflage animal detection algorithm can identify human brain cancer.

“This investigation is the first of its kind to apply camouflage animal transfer learning to deep neural network training on a tumor detection and classification task,” wrote lead author Arash Yazdanbakhsh, M.D., Ph.D., a research assistant professor in the Department of Psychological & Brain Sciences at Boston University, in collaboration with Faris Rustom, Ezekiel Moroze, Pedram Parva, and Haluk Ogmen.

Worldwide, brain and central nervous system cancers accounted for over 321,000 new cases and 248,500 deaths in 2022, according to a Global Cancer Observatory report by the World Health Organization’s International Agency for Research on Cancer.

Approximately 90,000 brain tumors are diagnosed each year in the United States, of which roughly 25,200 are cancerous according to the American Cancer Society’s Cancer Facts & Figures 2024; National Brain Tumor Society; Global Coalition for Adaptive Research. In 2024, there will be an estimated 25,400 new cases of brain and other nervous system cancers and 18,760 deaths in the U.S. according to the same report.

In the U.S., an estimated 1 million people live with a primary brain tumor and roughly 28 percent of all brain tumors are malignant, according to the National Brain Tumor Society (NBTS). The deadliest form of brain cancer, glioblastoma (GBM), makes up 50 percent of all primary cancerous brain tumors in America and has a median survival of eight months and a five-year relative survival rate of 6.9 percent per NBTS.

A brain tumor is the growth of abnormal cells in the brain tissue. A primary brain tumor is any tumor that originates in the brain or the brain area, as opposed to a metastatic brain tumor, which is a cancer that has spread to the brian from another part of the body.

Primary brain tumors can be subdivided into glial tumors or gliomas, and non-glial tumors. The human nervous system contains neurons and non-neuronal cells called glia. Neurons, also known as nerve cells or neurones, are excitable cells that transmit electrochemical impulses. The human brain consists of roughly 86 billion neurons, according to Vanderbilt University neuroscientist Suzana Herculano-Houzel, Ph.D., per her 2012 paper published in the Proceedings of the National Academy of Sciences of the United States of America (PNAS).

The word glia comes from the ancient Greek word γλία for “glue.” Within the central nervous system (CNS), the main types of glia include astrocytes, microglia, and oligodendrocytes; in the peripheral nervous system (PNS), satellite glial cells, enteric glia, and Schwann cells are examples of glia cells.

Glia cells are plentiful in the human brain. Neuroscientist Nicola Allen at the Salk Institute for Biological Studies in La Jolla, California, and neurobiology professor David A. Lyons at the University of Edinburgh in Scotland estimate that glia comprises around 50 percent of the central nervous system and plays a role in the formation and function of the central nervous system. An earlier estimate proposes that glial cells comprise 90 percent of brain cells according to a paper published in 2007 in The International Journal of Biochemistry & Cell Biology by University of California, Los Angeles, researchers Fei He and Yi E. Sun.

Not all brain tumors are deadly. There are over 150 distinct types of brain tumors that have been identified, according to the American Association of Neurological Surgeons (AANS). An estimated 27.9 percent, roughly one-third of all brain and central nervous system tumors in the U.S. are cancerous tumors, also known as malignant tumors, according to the American Brain Tumor Association.

What sets this current study apart is the incorporation of an AI transfer learning step from detecting camouflaged animals to finding brain tumors in MRI images, with an emphasis on explainability and transparency.

“Although camouflage animal detection and brain tumor classification tasks involve different images, there might be a parallel between an animal hiding through natural camouflage and a bundle of cancerous cells blending in with the surrounding healthy tissue,” the scientists wrote.

The researchers hypothesized that an AI network that was trained to spot animals in camouflage could be repurposed effectively to detect brain tumors from image data obtained noninvasively from magnetic resonance imaging (MRI) brain scans. In radiology, the two main types of MRI images are T1-weighted and T2-weighted. T1-weighted images highlight fat and are ideal for normal soft-tissue anatomy whereas T2-weighted images are ideal for spotting fluid and abnormalities such as tumors, trauma, and inflammation according to the 2023 Merck Manual.

The scientists repurposed an AI convolutional neural network (CNN) that was pretrained on identifying camouflaged animals into two AI models, one to classify T1-weighted MRIs, called T1Net, and the other, named T2Net, to classify T2-weighted MRIs.

The team deployed the explainable AI techniques of a feature visualization method called DeepDreamImage, image saliency mapping, and feature spaces.

The brain scan data for gliomas was mostly from the Cancer Imaging Archive (TCIA) of NIH National Cancer Institute and Kaggle public databases. Tumor categories included oligodendrogliomas, oligoastrocytomas, and astrocytomas. Data from normal MRIs of deidentified patient records from the Boston VA Healthcare System were also used to train the artificial neural networks as a control.

“T1Net and T2Net both had near perfect accuracies on normal brain images, with only 1-2 false negatives between both networks, demonstrating a strong ability to differentiate between cancerous and normal brains,” the scientists reported.

The transfer learning from the animal camouflage detection boosted the AI ability to classify brain tumors, especially astrocytomas. The transfer learning boosted AI model achieved a 92.2 percent accuracy for the T2-weighted MRI model, which outperformed the model without transfer learning.

“Our results demonstrate that this approach to deep neural network training is promising, specifically when using T2-weighted MRI data, as this model showed the greatest improvement in testing accuracy,” the researchers shared.

The researchers also report results that show that the qualitative XAI methods used enabled the visualization of what occurs during training of the AI using brain cancer MRI data, as well as traits associated with various tumor types. Through explainable AI methods, it was shown that the glioma classification decision-making process considered a tumor-specific feature-based methodology by the AI models.

The AI deep learning renaissance is enabling significant scientific advancements. Using AI to distinguish between cancerous and noncancerous brain tumors noninvasively has potential as an assistive tool for clinicians, oncologists, and radiologists in the future.

Copyright © 2024 Cami Rosso All rights reserved.