Scientists have combined machine learning with blood metabolite information to develop a new early diagnostic test capable of detecting ovarian cancer with 93% accuracy.
For more than three decades, doctors have been unable to perform highly accurate early diagnostic tests for ovarian cancer. Now, scientists at the Georgia Institute of Technology's Comprehensive Cancer Research Center (ICRC) have combined machine learning with blood metabolite information to develop a new test that can detect ovarian cancer in samples from the team's research group with 93 percent accuracy.
John McDonald, founding director of the ICRC and corresponding author of the study and professor emeritus in the School of Biological Sciences, explained that the new test is more accurate than existing tests for women with a clinical classification of normal than existing tests, and is particularly improved. Detection of early ovarian disease in this cohort.
The team's results and methods are detailed in a new article, "A Personalized Probabilistic Approach to Ovarian Cancer Diagnosis," published in the medical journal Oncology March 2024. Based on their computer model, the researchers developed what they thought would be a more clinically useful method for diagnosing ovarian cancer – a patient's individual metabolic profile that could be used to more accurately determine the probability of the presence or absence of the disease.
"This personalized, probabilistic approach to cancer diagnosis is more clinically informative and accurate than traditional binary (yes no) testing, and it represents a promising new direction for the early detection of ovarian cancer, as well as other cancers," McDonald said. ”
Silent killer
Ovarian cancer is often referred to as the silent killer because the disease is usually asymptomatic when it first appears and is often not detected until a later stage of development, when it is difficult to **.
McDonald explains that while the average five-year survival rate for patients with advanced ovarian cancer (even after **) is about 31%, if ovarian cancer is detected early and**, the average five-year survival rate will be more than 90%. "Clearly, there is a great need for accurate early diagnostic testing for this underlying disease. ”
Although the development of early detection tests for ovarian cancer has been vigorously pursued for more than three decades, the development of early, accurate diagnostic tests has proven difficult to achieve. MacDonald explains that because cancer starts at the molecular level, there are multiple possible pathways that can even lead to the same type of cancer.
"Because of this high level of molecular heterogeneity among patients, it is not possible to identify a single universal diagnostic biomarker for ovarian cancer, and for this reason, we chose to use a branch of artificial intelligence – machine learning – to develop another probabilistic approach to address the challenges of ovarian cancer diagnosis," McDonald said. ”
Metabolic profile
The study contributed to the study by Georgia Tech co-author Dongjo Ban, who explained, "It is well known that endpoint changes at metabolic levels reflect potential changes that work together at multiple molecular levels, so we chose metabolic profiling as the backbone of our analysis. ”
Co-author Jeffrey Skolnick adds, "The collection of human metabolites is a collective measure of cellular health, and by not arbitrarily selecting any subset in advance, it allows AI to figure out which are the key players in a particular cell. ”
Mass spectrometry can identify the presence of metabolites in the blood by detecting their mass and charge signatures. However, Ban says that the precise chemical composition of metabolites requires more extensive characterization.
Ban explains that because to date, less than 7% of the precise chemical composition of metabolites circulating in human blood has been chemically characterized, it is currently impossible to accurately pinpoint the specific molecular processes that affect an individual's metabolic profile.
However, the team recognized that even if the precise chemical composition of each metabolite is not known, the presence of different metabolites in the blood of different individuals can be included as features in the construction of accurate** models based on machine learning (similar to the use of individual facial features in building facial pattern recognition algorithms) even without knowing the precise chemical composition of each metabolite.
Thousands of metabolites are known to circulate in the human bloodstream, and they can be easily and accurately detected by mass spectrometry, combined with machine learning, to establish an accurate ovarian cancer diagnosis. Ban said.
A new approach to probability
The researchers developed their comprehensive approach, combining metabolomic signatures and a machine learning-based classifier, to build a diagnostic test with 93 percent accuracy when tested on 564 women from Georgia, North Carolina, Philadelphia, and Western Canada. 431 of the study participants were active ovarian cancer patients, while the remaining 133 women in the study did not have ovarian cancer.
Further research has begun to look at the possibility that the test will be able to detect very early stages of disease in women with no clinical symptoms, MacDonald said.
McDonald expects that in the clinical future, a person's metabolic profile is in the score range of the extremely unlikely to develop cancer and only needs to be monitored annually. However, if a person's metabolic score is in the range of the majority (e.g., 90%) of those who have been previously diagnosed with ovarian cancer, they may be monitored more frequently or may be referred immediately for advanced screening.
Hotspot Engine Program