Artificial intelligence (AI) computer vision is making significant inroads in digital healthcare. A new study published in IEEE Journal of Translational Engineering in Health and Medicine demonstrates how an AI method can predict therapy outcomes in patients with brain metastases—cancer that has spread to the brain from another organ—better than oncologists.
“Deep learning models have shown great promise in recognizing important and distinctive aspects of medical image data in various applications including cancer therapeutics,” wrote the researchers affiliated with the Sunnybrook Research Institute in Toronto, Canada, York University, and the University of Toronto who conducted the new study.
Metastatic brain cancer, also called secondary brain tumor, happen when cancer cells spread (metastasize) from their original location to the brain by traveling through the lymph system or bloodstream. The cells that form metastatic brain tumors may be from any organ and often come from cancers of the breast, lung, skin (melanoma), colon, intestines, kidney, thyroid, or ovaries. Up to 20 percent of cancer patients will develop a metastatic brain tumor according to Yale Medicine.
The symptoms of metastatic brain tumors may include constant headaches that get worse over time, seizures, emotional changes, personality changes, nausea, vomiting, memory issues, inability to move parts of the body such as an arm or leg, hearing issues, changes in vision, weakness, or numbness on one side of the body, balance issues, problems swallowing, sleepiness, and difficulty with speech comprehension or expression.
Radiation therapy is frequently used to treat brain metastases to manage the symptoms. A targeted radiation treatment called stereotactic radiosurgery or surgery may be used to treat those with just a few metastases.
“A noticeable proportion of larger brain metastases (BMs) are not locally controlled after stereotactic radiotherapy, and it may take months before local progression is apparent on standard follow-up imaging,” the researchers wrote. “This work proposes and investigates new explainable deep-learning models to predict the radiotherapy outcome for BM.”
To create and optimize the AI model, the researchers used image data from MRI (magnetic resonance imaging) scans of 124 brain metastases patients with 156 lesions who were treated with larger-dose radiation treatments given over a shorter period than conventional radiation called hypo-fractionated stereotactic radiotherapy. The deep learning model was tested with a different data set consisting of 25 patients with 40 lesions.
“Deep models, and especially convolutional neural networks (CNNs), can detect complex textural patterns in tissue, distinguish between malignant and benign cells, and possibly derive information from tumor images for therapy outcome prediction,” wrote the scientists. “Accordingly, the CNNs can potentially outperform the traditional radiomic models in diagnostic and prognostic applications for precision oncology by detecting patterns in medical images that are not captured by closed-form mathematical definitions of hand-crafted radiomic features.”
The researchers used BiT-HyperRule for hyperparameter selection and the models were trained using a stochastic gradient descent (SGD) optimization algorithm. The team created a novel transformer-convolutional network architecture that consists of residual learning (3D residual network) with a self-attention mechanism called CBAM (convolutional block attention module). The self-attention mechanism enables the AI algorithm to learn which areas of the images it should focus on more for processing and prediction.
“This study demonstrates the potential of self-attention-guided deep-learning features derived from volumetric MRI in radiotherapy outcome prediction for BM,” the researchers reported.
According to a statement by York University, the researchers tested different AI models with their best-performing model achieving an 83 percent accuracy, which is higher than human oncologists who on average achieve 65 percent accuracy in predicting radiotherapy failure in brain metastasis.
“The obtained results are promising and encourage future studies on larger patient populations,” the researchers concluded. With this successful proof-of-concept, the suggested follow-on research would be to use larger patient cohorts and multi-institutional data prior to developing a clinical trial in the future.
Copyright © 2022 Cami Rosso All rights reserved.