Read more

October 02, 2024
2 min read
Save

AI model predicts which ductal carcinoma in situ may progress to deadly breast cancer

You've successfully added to your alerts. You will receive an email when new content is published.

Click Here to Manage Email Alerts

We were unable to process your request. Please try again later. If you continue to have this issue please contact customerservice@slackinc.com.

Researchers have developed an AI model capable of using an inexpensive, easy-to-obtain chromatin image to determine the stage of ductal carcinoma.

The approach could help predict which DCIS cases will progress to invasive cancer, helping to avoid overtreatment.

Person touching AI
Researchers have developed an AI model capable of using an inexpensive, easy-to-obtain chromatin image to determine the stage of ductal carcinoma. Image: Adobe Stock.

DCIS — a preinvasive tumor — accounts for about one-quarter of breast cancer diagnoses.

An estimated 30% to 50% of DCIS cases progress to invasive breast carcinoma. However, clinicians have difficulty determining DCIS type and stage, making it difficult to determine which cases will progress, meaning many patients undergo immediate treatment unnecessarily.

G.V. Shivashankar, PhD, full professor in the department of health sciences and technology at ETH Zurich in Switzerland, and colleagues created a model that demonstrated the morphological state and organization of cells in tissue can be used to determine DCIS stage.

“The use of genome packing as a readout, combined with AI for cancer staging, is a new concept that could open interesting avenues for collaboration between pathologists and AI models,” Shivashankar told Healio. “If the model is proven effective, we hope it could aid in identifying particular tissue regions for deeper pathological inspection.”

Techniques such as multiplexed staining or single-cell RNA sequencing are sometimes used to determine the stage of DCIS in tissue samples; however, these tests are generally too costly for widespread use.

Shivashankar and colleagues previously determined chromatin staining could provide equally valuable information at a much lower cost versus single-cell RNA sequencing.

In the current study, researchers theorized that combining chromatin staining with a carefully devised machine-learning model could determine cancer stage as effectively as more expensive approaches.

They created a large-scale tissue microarray dataset consisting of chromatin images of 560 samples from 122 women with DCIS. Because these images are easily acquired, the research team has amassed one of the largest databases of its kind.

Investigators used the data to train and assess an AI model that generates clusters of cells in similar morphologic states and recognizes eight cell states that are markers of DCIS.

Some cell states are more suggestive of invasive cancer than others, and the model calculates the proportion of cells in each state in a tissue sample.

“Based on this imaging data, our AI model detects how our genomes are packed in each cell, and how cells are organized spatially within the tumor microenvironment,” researcher Xinyi Zhang, a graduate student in MIT’s department of electrical engineering and computer science, told Healio. “It extracts features of the neighborhood of cells as well as the genome packing to predict the different stages of DCIS.”

The researchers independently cross-validated the model with a senior pathologist, Zhang said. In many cases, the model’s results and those of the pathologist aligned. In less straightforward cases, the model could provide information about the organization of cells in a tissue sample that a pathologist could use to guide treatment decisions.

The model could be used in other types of cancer and also is being evaluated for possible se in neurodegenerative conditions, according to investigators.

“In order for this model to be used in the clinic, follow-up large-scale clinical trials would be needed, where patients are analyzed prospectively using the combination of imaging and AI model proposed in our paper,” researcher Caroline Uhler, PhD, Andrew and Erna Viterbi professor of engineering at MIT and a core institute member of Broad Institute of MIT and Harvard, told Healio. “If such large-scale clinical trials are successful, regulatory board approvals would need to be sought before widespread clinical use.”

For more information:

Caroline Uhler, PhD, can be reached at cuhler@mit.edu.