Stuart Dalrymple1
1 Franklin.ai, Sydney, NSW, Australia
Abstract for Conference Poster at 36th European Congress of Pathology, 7 - 11 September 2024, Florence, Italy
The integration of artificial intelligence (AI) in pathology possesses great potential, however widespread adoption remains limited due in part to limited digital pathology infrastructure. Here we utilised an AI model integrated within a microscope-based system in prostate cancer histopathology.
We analysed a dataset comprising 440 images of prostate needle core biopsy (NCB) specimens capturing 26 clinical findings. Using a previously trained and validated model, we evaluated separately both whole slide images (WSI) scanned, and camera integrated microscope acquired image data of the same slide. We explored both quantitative (standard performance metrics) and qualitative (visualisation) analyses to assess comparative performance.
Our preliminary results demonstrate the feasibility of our approach in integrating AI technologies to microscope-based pathology. Microscope acquired images assessed by our model demonstrated strong classification precision for Acinar Gleason pattern 3, 4 and 5 along with other focal histological markers including intraductal carcinoma (Precision > 0.7 for all).These results were comparative or better than the same scanned WSIs assessed by our model highlighting that despite variations in image acquisition modalities, the AI model exhibited robust performance across both WSI and microscope image datasets.
We have shown that our model was robust in classification of focal markers of prostate cancer by incorporating AI techniques into traditional microscopy diagnostic methods. Importantly, we find we can add similar diagnostic and reporting value across Gleason grades (3-5) and prognostic factors such as perineural invasion, which form the basis of most prostate cancer reporting. There is potential for laboratories to use traditional microscopy combined with our model which could lower the barrier to AI adoption.
DOI: https://doi.org/10.1007/s00428-024-03880-y
The following was accepted as an ePoster for ECP 2024
The results presented in this abstract are based on an early version of our AI model therefore outcomes reported here are subject to change as the model undergoes further development and validation.
Interested in learning more? Get in touch