Amsterdam UMC uses analytics and AI to increase speed and accuracy of tumor evaluations
The application of artificial intelligence (AI) is gaining traction in oncologic care, and Amsterdam UMC is leading the way by using computer vision and predictive analytics to better identify cancer patients who are candidates for lifesaving surgery.
With one of Europe’s largest academic oncology centers, Amsterdam UMC strives for every patient to contribute to the care of the next patient. This is done by collecting enormous amounts of data on each patient, including biomarkers, DNA and genomic data.
“Our opportunity is to use AI to help us with our ever-growing data volumes,” says Dr. Geert Kazemier, Professor of Surgery and Director of Surgical Oncology at Amsterdam UMC.
His search for a robust analytics platform led Kazemier to SAS, kicking off a partnership that has furthered the science of using AI to evaluate liver tumors pre- and post-systemic therapy. Additionally, the SAS platform gives thousands of cancer researchers at Amsterdam UMC access to cutting-edge analytics to improve research and collaboration.
AI will help us save lives ... I’m absolutely sure about that. Dr. Geert Kazemier Professor of Surgery and Director of Surgical Oncology Amsterdam UMC
Human limitations in tumor assessments
Colorectal cancer is the third-most common cancer worldwide, and it spreads to the liver in about half the patients. Kazemier, who specializes in liver surgery, says the best way to treat this type of cancer is to remove it. But some tumors are too large to be removed, and these patients must undergo systemic therapy, such as chemotherapy to shrink the tumors.
After a period of treatment, tumors are manually evaluated using computerized tomography (CT) scans. At that time, medical professionals can see if a tumor shrunk or changed in appearance. How a tumor reacts to systemic therapy determines whether lifesaving surgery is possible or if a different chemotherapy regimen is necessary.
This manual approach presents many challenges.
Evaluating tumors is a time-consuming process for radiologists. And for each CT scan, typically only the two largest tumors are measured – possibly leaving vital clues hidden in the remaining tumors, if a patient has more than two. Furthermore, the manual assessment is prone to subjectivity, which results in variation of response evaluation among radiologists.
3D evaluation is also lacking, according to Kazemier. “A tumor might shrink but not symmetrically. This is difficult to quantify with the human eye,” he explains. In other instances, tumors might change appearance, indicating there is less blood running through the metastasis – a positive sign of systemic therapy effectiveness, which is also hard for humans to detect.
Then there is human error. Unfortunately, due to human limitations, radiologic errors happen, and a misdiagnosis can subject a patient to life-threatening risks, such as unnecessary surgery or chemotherapy.
Amsterdam UMC – Facts & Figures
scientific articles published annually
of Europe’s largest academic oncology centers
AI detects tumors faster and more accurately than humans
Together with SAS, Amsterdam UMC is transforming tumor evaluations with AI. It uses computer vision technology and deep learning models in SAS Visual Data Mining and Machine Learning to increase the speed and accuracy of chemotherapy response assessments. Data scientists also take advantage of the SAS Deep Learning With Python (DLPy) API to create deep learning models. Capabilities like automatic segmentation help doctors quickly identify changes in the shape and size of tumors and note their color.
“We’re now capable of fully automating the response evaluation, and that’s really big news,” Kazemier says. “The process is not only faster but more accurate than when it’s conducted by humans.”
The project started by training a deep learning model with data from 52 cancer patients. Every pixel of 1,380 metastases was analyzed and segmented. This taught the system how to instantly identify tumor characteristics and share vital information with doctors.
Prior evaluation methods limited what doctors could see, but the AI models provide total tumor volume and a 3D representation of each tumor, allowing doctors to more accurately determine whether lifesaving surgery is viable or a different treatment strategy should be chosen.
“AI will help us save lives ... I’m absolutely sure about that,” Kazemier says.
Advanced analytics for cancer researchers
Outside the clinic, the SAS platform is also available to more than 1,100 Amsterdam UMC cancer researchers to enhance their research efforts. SAS Visual Analytics allows them to quickly spot hidden trends, while SAS Visual Statistics provides a powerful tool to perform advanced analytics and predictive modeling.
Additionally, SAS Viya supports the automatic translation of raw images to objective metrics in a clinical setting. Such automation will save radiologists a lot of time, while reducing the number of dangerous false negatives and false positives.
By running these solutions on SAS Viya, Amsterdam UMC gives researchers an open analytics platform to collaborate and obtain innovative results faster. Now, biologists, doctors, medical students and even business analysts working to improve the patient journey can benefit from analytics regardless of their data skills or coding language preference.
“There are a lot of people working with the SAS platform who do not have analytic or data science training,” Kazemier says. “This is the next phase of analytics for us, and I see tremendous opportunities ahead.”
For Kazemier, AI technology must be transparent and open if it’s going to revolutionize healthcare. “If you create algorithms to help doctors make decisions, it should be explainable what that algorithm is actually doing,” he says. “Imagine if an algorithm came up with something bad for the patient and the doctor follows it. What’s the effect of that? To err is not only human.”
A critical factor in the ongoing deployment of analytics within clinical settings is to establish an end-to-end auditable and transparent process for decision management in health care. SAS provides a summary of how each analysis is performed, making it easier for doctors to track their models and algorithms. This improved collaboration between human and machine builds more trust in AI. This level of transparency attracted Kazemier to SAS.
“We needed an explainable model while still maintaining a high level of learning performance,” he says. “SAS was the most trustworthy solution we found.”
Looking forward, Kazemier sees a bigger role for AI at Amsterdam UMC.
“In the future, we may be able to predict the outcome of surgery and overall patient survival,” he says. “While we are currently using AI technology with colorectal liver cancer patients, AI has the potential to be used in assessing many solid tumor types, including breast and lung cancer. We have only touched the tip of the iceberg.”