Background & objectives: Artificial Intelligence (AI) can be trained to recognise complex patterns and morphology within digitised histopathology specimens and use them to aid clinical reporting. These patterns may be already known, where the pathologist trains the AI model. However, cancer is a complex disease and the tissue may harbour undiscovered but clinically significant morphological patterns. AI can also be used to identify and report such novel features without human bias and error. This talk will demonstrate how the two above methodologies were successfully used in colorectal (CRC) and bladder cancer examples and how they will be used for the automatic reporting of gynaecological specimens in the iCAIRD initiative. Methods: Machine learning workflows were designed to analyse unbiasedly extracted data from the automated analysis of immunofluorescence labelled urine cytology (n=624) and whole slide CRC specimens (n=173). H&E labelled whole slides of CRC were analysed using AI after both pathology trained feature recognition (n=650) and without any human direction (n=75). Results: Bladder cancer diagnosis was reported from urine cytology samples with 95% sensitivity and 70% specificity. CRC survival was predicted with an AUROC of 0.94 in the immunofluorescence labelled cohort using 123 automatically extracted features and with>95% accuracy in the H&E stained digitised slides when applying no human input.Conclusion:The use of AI allows clinically significant features to be reported with or without human training and error. This work demonstrates AI’s ability to automatically report clinical specimens and how the technology can track toward clinical translation such as with the iCAIRD initiative.
|Number of pages
|Published - Sept 2019
|31 st European Congress of Pathology - Nice Acropolis Convention Centre, Nice, France
Duration: 7 Sept 2019 → 11 Sept 2019
|31 st European Congress of Pathology
|7/09/19 → 11/09/19