Ductal carcinoma in situ (DCIS) is a type of preinvasive tumor that can progress to a deadly form of breast cancer, accounting for about 25 percent of all breast cancer diagnoses.
A new machine-learning model has been developed to identify the stage of disease in DCIS, which could help clinicians avoid unnecessary overtreatment of patients. The model, created by researchers from MIT and ETH Zurich, analyzes breast tissue images to determine the different stages of DCIS based on cell state and arrangement in the tissue sample.
By utilizing a large dataset of tissue images, the researchers trained and tested the AI model, which showed promising results in predicting the stage of DCIS when compared to conclusions drawn by a pathologist.
This model could potentially assist clinicians in streamlining the diagnosis process for simpler cases, allowing more time to focus on cases where the progression of DCIS to invasive cancer is less clear.
The researchers emphasize the importance of considering the spatial organization of cells in diagnosing DCIS and are looking to conduct prospective studies to further validate the model’s effectiveness in clinical settings.
Combining Imaging with AI
Current techniques to determine the stage of DCIS in tissue samples are costly and not widely accessible. In this study, researchers combined a cost-effective imaging technique with machine learning to provide valuable information about cancer stage.
By creating a dataset of tissue sample images and training an AI model to analyze the state and organization of cells, the researchers were able to identify key markers of DCIS and improve the accuracy of cancer stage prediction.
Organization Matters
The researchers found that not only the proportion of cells in different states but also the arrangement of cells are crucial in determining cancer stage. By incorporating spatial organization into their model, they significantly enhanced its accuracy.
This focus on spatial organization proved to be a key factor in the model’s success, providing valuable insights for pathologists in decision-making processes.
This adaptable model shows promise for potential application in other types of cancer and neurodegenerative conditions, with ongoing exploration by the researchers.
The research was supported by various institutions and foundations, highlighting the collaborative effort to advance AI applications in healthcare and cancer research.
