A deep learning approach to case prioritisation of colorectal biopsies
- Ciara D White 1,2, Runjan Chetty 2, John Weldon 2, Maria E Morrissey 2, Rob Sykes 2, Corina Gîrleanu 1, Mirko Colleuori 2, Jenny Fitzgerald 2, Adam Power 2, Ajaz Ahmad 2, Seán Carmody 2, Pierre Moulin 2, Donal O'Shea 2, Muhammad Aslam 3,4, Mahomed A Dada 3,5, Maurice B Loughrey 3,6, Martine C McManus 3, Klaudia M Nowak 3,7, Kristopher McCombe 4, Sinead Hutton 1, Máirín Rafferty 2, Niall Mulligan 1
- Ciara D White 1,2, Runjan Chetty 2, John Weldon 2
- 1Department of Histopathology, Mater Misericordiae University Hospital, Dublin, Ireland.
- 2Deciphex, DCU Alpha Innovation Campus, Dublin, Ireland.
- 3Diagnexia, Exeter, UK.
- 4Betsi Calawaladar NHS Health Board, Wales, UK.
- 5Royal Derby Hospital, Derby, UK.
- 6Centre for Public Health, Department of Cellular Pathology, Queens University Belfast, Belfast Health and Social Care Trust, Belfast, UK.
- 7University Health Network, Toronto, Canada.
- 0Department of Histopathology, Mater Misericordiae University Hospital, Dublin, Ireland.
Related Experiment Videos
Contact us if these videos are not relevant.
Contact us if these videos are not relevant.
View abstract on PubMed
Summary
This summary is machine-generated.A new artificial intelligence (AI) model effectively detects abnormal colorectal histology, including cancer and dysplasia. This AI tool assists pathologists by prioritizing biopsies, improving diagnostic workflows.
Area Of Science
- Digital Pathology
- Artificial Intelligence in Histopathology
- Colorectal Cancer Screening
Background
- Accurate and timely diagnosis of colorectal histology is crucial for patient outcomes.
- Traditional histopathological analysis can be time-consuming and subject to inter-observer variability.
- Artificial intelligence (AI) offers potential to enhance diagnostic accuracy and efficiency in pathology.
Purpose Of The Study
- To develop and validate a weakly supervised AI model for detecting abnormal colorectal histology.
- To prioritize colorectal biopsies based on clinical significance, distinguishing between neoplastic and non-neoplastic findings.
- To assess the AI model's integration into a digital pathology workflow.
Main Methods
- A weakly supervised deep learning model, Triagnexia Colorectal, was trained on 24,983 digitized H&E-stained whole slide images.
- The model was evaluated by multiple pathologists in a simulated digital pathology environment.
- An AI application with a graphical user interface was developed to streamline decision-making.
Main Results
- The AI model achieved high performance in validation cohorts, with micro-average specificity of 0.984 and sensitivity of 0.949 on the first cohort (n=100).
- A secondary multi-institutional cohort (n=101) demonstrated comparable results with micro-average specificity of 0.978 and sensitivity of 0.931.
- Pathologists reported positive feedback on the AI tool's accuracy, utility, and ease of integration.
Conclusions
- A high-performing AI triage model for colorectal biopsies has been successfully developed.
- The AI model can be integrated into routine digital pathology workflows.
- This AI tool assists pathologists in prioritizing cases and identifying significant abnormalities like dysplasia and cancer.
Related Experiment Videos
Contact us if these videos are not relevant.
Contact us if these videos are not relevant.