AI Outperforms Pathologists in Predicting Cancer Spread to Brain

Geralt/Pixabay

One of the most important uses of artificial intelligence (AI) machine learning is in health care and precision medicine. A new scientific study published on Monday in The Journal of Pathology reveals that AI deep learning can predict if a patient’s non-small cell lung cancer (NSCLC) will spread to the brain within five years of diagnosis better than expert pathologists.

“Ours is the first study to our knowledge that has explored the prediction of future brain metastases based on the histology of the primary tumor in early-stage NSCLC,” wrote corresponding author Richard Cote, a professor, department chair, and pathologist-in-chief at the Barnes-Jewish Hospital at the Washington University School of Medicine in St. Louis, along with first author Haowen Zhou at Caltech in collaboration with Caltech researchers Steven (Siyu) Lin and Simon Mahler, and Washington University School of Medicine researchers Ramaswamy Govindan, Sid Rawal, Alexander Wein, Jon Ritter, Chieh-yu Lin, Cory Bernadt, and Mark Watson.

Cancer is the second-most-common cause of death worldwide, according to Our World in Data. In 2019, cancer caused 10 million deaths annually, second only to heart diseases (heart attacks, stroke, other cardiovascular diseases) at 18.5 million deaths each year.

An estimated one in every five people worldwide will develop cancer during their lifetime, and by 2050 there will be more than 35 million new cancer cases, according to a 2022 report by the International Agency for Research on Cancer (IARC) at the World Health Organization. Globally, lung cancer is the most common cancer and the leading cause of cancer deaths, per the IARC. Specifically, out of the total new cases worldwide, the IARC reports that lung cancer is the most common, with female breast cancer as a close second, followed by colorectal cancer, prostate cancer, and stomach cancer.

A majority of U.S. lung cancer cases, approximately 80 to 85 percent, are NSCLC, and the remaining are small cell lung cancer (SCLC) cases according to the American Cancer Society (ACS). Three cancers that may originate in or near the lungs but are not classified as lung cancer include pleural mesothelioma, a rare cancer where an estimated 70 percent of cases are due to asbestos exposure that affects the lining of the lungs and chest; sarcoma, a rare cancer that develops in the bone and/or soft tissues; and lymphoma, a blood cancer that originates in the white blood cells, also known as lymphocytes, in the body’s lymphatic system, according to the Cleveland Clinic.

Brain Metastases

Brain metastases are one of the most common complications of cancer and happen when cancer cells that originated from anywhere else in the body spread to the brain. Patients with lung cancer, breast cancer, and melanoma are at the greater risk of brain metastases, according to the Epidemiology of Brain Metastases study by Patrick Wen et al. published in Neuro-oncology. In addition to the top-three most-likely cancers to spread to the brain (lung cancer, breast cancer, and melanoma), brain metastases also often happen to patients with renal cell carcinoma, colon cancer, and gynecologic cancers according to MD Anderson.

Cancer spread to the brain can happen in almost half of those diagnosed with stages I to III NSCLC, and currently there is no dependable molecular nor histopathologic (examination of tissues and cells obtained via biopsy or surgery under a microscope) method to predict brain metastases, according to the Caltech and Washington University research team.

Using an AI Deep Learning Classifier

Can the predictive capabilities of artificial intelligence (AI) machine learning provide a reliable way to identify the risk of NSCLC patients developing brain metastases? To answer this question, the scientists developed an AI deep learning (DL) classifier consisting of a ResNet-18 convolutional neural network (CNN) that was pretrained on data from ImageNet.

Overall, the study used tissue images from 158 patients with stage I to III NSCLC diagnosed and treated by the Washington University School of Medicine that were followed until metastasis or for half a decade or more. The AI model was trained using actual diagnostic H&E-stained tumor tissue slides from 118 stage I to III NSCLC patients, out of which 45 eventually developed brain metastases.

H&E stands for hematoxylin and eosin, which are the two dyes used for the most common method of staining in histology, the microscopic study of cells and tissues. Hematoxylin consists of an extract from the logwood tree (Haematoxylum campechianum), and eosin is an acid dye derived synthetically according to the Mayo Clinic Proceedings. Ever wonder, once a tissue sample is obtained either through a biopsy or surgery, exactly what is the process that happens behind the scenes to determine if the sample is cancerous? To prepare the glass slides for microscopy, samples are fixed (preserved), processed, and embedded in paraffin to be hardened up for sectioning. As a final step, the sample is stained to enhance visibility, given most cells are clear and colorless according to The Histology Guide by the University of Leeds.

Artificial Intelligence Essential Reads

AI's Turing Test Moment

AI's Quest for a Grand Unification Theory

For this study, the researchers tested their deep learning algorithm on the histological slide images of 40 different stage I to III NSCLC patients, of which half had developed brain metastases. The AI performance was compared to that of four expert human pathologists. The AI had a prediction accuracy of 87 percent, which was much higher than the 57.3 percent accuracy average among the pathologists.

“Here we demonstrate how a DL network can be effectively trained on digital images from routine H&E-stained NSCLC tumor tissue slides to predict brain metastatic progression within 5 years of the initial diagnosis and, importantly, accurately identify those cases that do not progress after 5 or more years of follow-up,” the scientists concluded.

Copyright © 2024 Cami Rosso. All rights reserved.