Using AI-based image analysis to predict ovarian tumor outcome

Publication on utilizing an AI model to accurately classify high grade serous carcinoma into outcome groups using tumor morphology alone.
Written by Aiforia

High-grade extrauterine serous carcinoma (HGSC) is an aggressive tumor with high rates of recurrence, mainly found in ovaries. These tumors have significant genetic and morphologic heterogeneity and are notable for their diversity and unpredictability in outcome. Treatment methods and research for these tumors have developed very little in the last few decades. Standard therapy includes debulking surgery and platinum-based chemotherapy, while overall 5-year survival rates remain below 50%.

Development of artificial intelligence (AI) based image analysis has improved prediction and identification of underlying molecular alterations in other tumor types, such as colorectal, lung, and breast carcinomas. These AI models have shown to perform as well as expert pathologists and offer increased precision in bias-prone tasks such as metastasis detection. This suggests a promising new treatment model for HGSC tumors as well.

There is an intrinsic difference in HGSC tumors that are platinum-based chemotherapy sensitive and resistant. However, this variation is currently not prospectively identifiable by pathologists. The authors of the recent publication in Nature have proposed that some indication of this underlying difference is detectable in the tumor morphology with the use of AI for differentiation of these two groups of patients.

“We used AI for this project because this tumor type (high grade serous carcinoma of the ovary) is a little unusual. When pathologists look at these slides, unlike for nearly every other carcinoma, we provide very little (or no) information beyond the diagnosis itself. At the same time, there are many different morphologies to look at! We hypothesized that AI might be able to identify something the human eye has not yet recognized.”
Anna Laury, MD, Clinical researcher at University of Helsinki

Blog Header Images (1)-1

Using a patient cohort reflecting the extremes of treatment response, the researchers developed an AI model with Aiforia Create to differentiate between good outcome and poor outcome tumors in whole slide images. AI has the potential to accurately classify high grade serous carcinoma into outcome groups using tumor morphology alone with significant benefits to treatment methods and survivability predictions for patients.

Read the full publication here.
Laury AR, Blom S, Ropponen T, et al. Scientific Reports, 2021