New research demonstrates that deep learning systems can help physicians make clinical decisions with highly accurate estimates of how long a patient is likely to live after being diagnosed with a disease. Researchers at Penn State Great Valley School of Graduate Professional Studies report using deep learning, a subset of machine learning, to develop several lung cancer survival prediction models.
In a new study, the deep learning models outperformed traditional machine learning in terms of cancer survival classification accuracy (71% vs. 61%) and regression approaches (13.5% vs.14.87%). The researchers used lung cancer data from the Surveillance, Epidemiology, and End Results (SEER) registry to build the cancer survival period prediction models, which were compared across three deep learning architectures-artificial neural networks, convolutional neural networks, and recurrent neural networks-in addition to traditional machine learning.
The researchers also evaluated the importance of data-referred to as features in machine learning-to investigate model interpretability and factors important in cancer survival prediction. The new deep learning models "contribute to identifying an approach to estimate survivability that is commonly and practically appropriate for medical use," the researchers concluded in an article detailing the study results in the International Journal of Medical Informatics (2021; https://doi.org/10.1016/j.ijmedinf.2020.104371).
"Artificial intelligence and machine learning will help transform the health care industry," said the study's corresponding author, Robin G. Qiu, PhD, Professor of Information Science in the Institute for Computational and Data Sciences (ICDS) and Professor-in-Charge of Information Science at Penn State Great Valley. "Deep learning approaches can serve as a basis for developing health care analytics models, or tools that will support health care professionals and patients to plan medical resources, treatment, and care strategy."
Traditional machine learning uses algorithms to separate data, self-educate, and make decisions based on what it has learned from that data. Deep learning uses artificial neural networks based on the function of the human brain layered in a sophisticated architecture of cells-hence the term "deep."
Deep learning neural networks identify recurring patterns in complex datasets using a machine learning algorithm that learns from associations it makes between data, itself, and the labels scientists use to describe data examples. This enables the extraction of hidden data from complex datasets.
Study Details
In the Penn State study, the researchers used the ICDS Roar supercomputer to analyze 800,000 to 900,000 patient records in 150 fields in the SEER cancer registry for the study-a task they said would have been virtually impossible for a team of humans to complete.
"Deep learning models significantly outperform traditional machine learning models when a large number of patients' longitudinal medical records are available," Qiu said.
Machine learning has been used to detect cancer via imaging, such as computed tomography (CT) scans; at least a dozen methods have been FDA-approved for use in radiology. But although detection, grading, and subtyping tumor tissues automates workflows, they don't necessarily impact clinical decision-making (Br J Cancer 2021; https://doi.org/10.1038/s41416-020-01122-x).
Deep learning techniques, however, such as inference of molecular features, prediction of therapy response, and also survival period prediction, can influence clinical decisions. While emerging deep learning concepts go even further to include the possibility of extracting biomarkers from histology images.
"The applicability of machine learning in medicine makes sense, especially given the large datasets inherent to humans and their diseases," said Jack Jacoub, MD, a medical oncologist and Medical Director of MemorialCare Cancer Institute at Orange Coast Medical Center in Fountain Valley, Calif. "It isn't surprising that a machine can accurately predict prognosis when all data is entered and analyzed, and probably becomes more efficacious and accurate with time as it learns previous events and outcomes.
"But, I would say that, more important than predicting prognosis, which oncologists are generally very proficient at, it would be extremely welcomed if machine learning can help choose a therapy given the multiple options that exist for most cancers now based on patient, tumor, and drug characteristics which is being studied at the moment," Jacoub continued.
"Indeed, the researchers highlight a very important and evolving area in medical informatics. And most believe it's only a matter of time until doctors are aided by such technology, which will further refine the concept of precision or tailored medicine," he added, referring to the Penn State Great Valley study specifically.
Osita Onugha, MD, thoracic surgeon and Assistant Professor of Thoracic Surgical Oncology at Saint John's Cancer Institute at Providence Saint John's Health Center in Santa Monica, Calif, agreed. Deep learning and artificial intelligence in medicine is the future of health care, he asserted.
"I think this is just beginning to scratch the surface of what we can do with artificial intelligence. Some of the most intriguing and interesting questions will be how we can use machine learning to predict lung cancer from a CT scan," Onugha said.
"Oftentimes, when reviewing CTs, there are lesions that we do not know if they represent a malignancy, so we watch them for a period of time because they are in the 'gray' area. However, using artificial intelligence, we can better delineate whether these gray area lesions are benign or cancer," he explained.
"Giving physicians a predictive model would allow us to be more efficient and effective in evaluating certain lesions," Onugha stated. "In addition, this could prevent unwanted surgeries if the model could predict with reasonable certainty that a lesion is benign. This would be an effective tool for physicians caring for lung cancer patients."
Meanwhile, back at Penn State Great Valley, Qiu said the research team will continue to build deep learning survival period prediction models for other types of diseases.
"We focus on predicting patients' future disease risks when discharged in terms of their probability of readmissions. Two different answers could be provided-when they will be readmitted and what diseases they might have when readmitted," she said. "We have applied a similar approach to predict the progression of Alzheimer's disease for patients. The outcomes are surprisingly good."
Chuck Holt is a contributing writer.