Artificial intelligence (AI) was used to predict response of high-grade extrauterine serous carcinoma (HGSC) to adjuvant platinum chemotherapy.
For the first time, researchers used artificial intelligence (AI) to predict response of high-grade extrauterine serous carcinoma (HGSC), a highly aggressive cancer with frequent chemotherapy resistance and a poor survival rate, to adjuvant platinum chemotherapy.
As explained in Scientific Reports, there are currently no validated tissue-based prognostic or predictive markers for primary platinum-based treatment in use for HGSC, a type of ovarian cancer; about 75% of ovarian cancer deaths are from HGSC. Typically presenting at stage III or stage IV, the 5-year survival rate for HGSC is less than 50% and about 25% in advanced disease.
These tumors are marked by genetic and morphologic heterogeneity and outcomes are uncertain, even when patients appear to have favorable characteristics, such as younger age and lower stage. About 15% of patients can survive for a decade or more. Standard treatment (debulking surgery and platinum-based chemotherapy) hasn’t changed for this disease in decades, and despite the arrival of poly ADP ribose polymerase inhibitors, treatment response is still unpredictable.
Patients with a progression-free interval (PFI) of less than 6 months are considered platinum resistant, while those with a PFI beyond 12 months are considered platinum sensitive; only 25% of women have a PFI of greater than 18 months.
The researchers, from Finland, said they believed the differences that exist in HGSC tumors—which ones are refractory to therapy immediately and which ones have later resistance—could be detectable using AI. The study aimed to see if a weakly supervised convolutional neural network trained on whole slide images (WSI) can accurately classify HGSC into outcome groups using the tumor morphology on the images. WSI refer to conventional slides that are scanned to create digital files.
The HGSC patient cohort (the test set) included 30 women with similar presentations who experienced very different treatment responses (PFIs of either ≤ 6 months or ≥ 18 months and 205 slide images).
In 3 steps, the neural network was trained to identify morphologic regions (digital biomarkers) that are highly associated with one or the other treatment response group. The classifier was tested using a separate 22-slide test set, and 18 of 22 slides were correctly classified.
The method was able to discern extremes in patient response to primary platinum-based chemotherapy with high sensitivity (73%) and specificity (91%).
The authors said the proof-of-concept results are novel since it is the first time prospective prognostic information is identified specifically within HGSC tumor morphology. The finding is important since unlike other tumor types, “pathologist evaluation of HGSC tumor tissue provides almost no predictive or prognostic information beyond the diagnosis itself,” the authors said.
Reference
Laury AR, Blom S, Ropponen T, Virtanen A, Carpén OM. Artificial intelligence-based image analysis can predict outcome in high-grade serous carcinoma via histology alone. Sci Rep. Published online September 27, 2021. doi:10.1038/s41598-021-98480-0
New Research Challenges Assumptions About Hospital-Physician Integration, Medicare Patient Mix
April 22nd 2025On this episode of Managed Care Cast, Brady Post, PhD, lead author of a study published in the April 2025 issue of The American Journal of Managed Care®, challenges the claim that hospital-employed physicians serve a more complex patient mix.
Listen
Higher Weight-Adjusted Waist Index Tied to Greater Mortality Risk in Patients With Osteoarthritis
April 23rd 2025Researchers consider the weight-adjusted waist index a more precise predictor of mortality risk in patients with osteoarthritis than traditional obesity measures, like body mass index.
Read More
Personalized Care Key as Tirzepatide Use Expands Rapidly
April 15th 2025Using commercial insurance claims data and the US launch of tirzepatide as their dividing point, John Ostrominski, MD, Harvard Medical School, and his team studied trends in the use of both glucose-lowering and weight-lowering medications, comparing outcomes between adults with and without type 2 diabetes.
Listen