This system uses multi-scale pyramidal pre-trained CNN to analyze both local and global pathology markers extracted from digital pathology images. Following training, the model performed highly, with an accuracy rate of 94.25%, a sensitivity rate of 94.47%, and a specificity rate of 94.03% in the testing dataset. ShuffleNet also outperformed three existing AI models for digital pathology grading. Shalata et al. subsequently tested the model’s performance against WHO classifications that characterize a tumor as high-grade if over 5% of a tissue slide is high-grade, regardless of subtype. On a low-grade tumor slide, the AI model misclassified 5 out of 1,132 patches as high-grade, equivalent to 0.004% of all the patches and less than the 5% standard. Similar results were obtained with high-grade tumors.
The AI model presented in this study showed superior performance to previous models. The application of ShuffleNet can enhance accuracy, especially in low-resource settings where trained genitourinary pathologists may not be available. This can allow for improved patient management and treatment outcomes. Further studies are needed to test AI systems in clinical settings prospectively and identify and address barriers to implementing AI in these settings.
Written by: Bishoy M. Faltas, MD, Director of Bladder Cancer Research, Englander Institute for Precision Medicine, Weill Cornell Medicine
References:
- Shalata AT, Alksas A, Shehata M, et al. Precise grading of non-muscle invasive bladder cancer with multi-scale pyramidal CNN. Sci Rep. 2024;14(1):25131. Published 2024 Oct 24. doi:10.1038/s41598-024-77101-6