AI Framework for Detecting Cancer Cell Division
Author Information
Author(s): Shen Zhuoyan, Simard Mikaël, Brand Douglas, Andrei Vanghelita, Al-Khader Ali, Oumlil Fatine, Trevers Katherine, Butters Thomas, Haefliger Simon, Kara Eleanna, Amary Fernanda, Tirabosco Roberto, Cool Paul, Royle Gary, Hawkins Maria A., Flanagan Adrienne M., Collins-Fekete Charles-Antoine
Primary Institution: University College London
Hypothesis
Can a deep learning framework improve the detection of mitotic figures in cancer pathology?
Conclusion
The study successfully developed a high-performance AI model that detects mitotic figures in various cancer types using the largest dataset of its kind.
Supporting Evidence
- OMG-Net achieved an F1 score of 0.84 in detecting pan-cancer mitotic figures.
- The study created the largest dataset of mitotic figures, combining in-house and open-source data.
- Integration of canine data improved detection performance in human tumors.
- Manual curation of masks significantly enhanced model accuracy.
- Active learning was employed to refine the dataset and improve model performance.
Takeaway
This study created a smart computer program that helps doctors find dividing cancer cells in tissue samples, making it easier to diagnose and treat cancer.
Methodology
The study used a two-stage framework called OMG-Net, which combines a segmentation model with a classification model to identify mitotic figures in digitized images.
Potential Biases
Inter-observer variability among pathologists may affect the accuracy of the model's training data.
Limitations
The model's performance varied across different tumor types, particularly lower for neuroendocrine tumors due to limited data and quality issues.
Participant Demographics
The dataset included mitotic figures from both human and canine specimens.
Statistical Information
P-Value
0.001
Statistical Significance
p<0.001
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website