Artificial Intelligence for Prostate Cancer Diagnosis and Prognostication
||Artificial Intelligence for Prostate Cancer Diagnosis and Prognostication|
||Martin Eklund <firstname.lastname@example.org>|
||2022-12-01 – 2023-06-01|
Digital pathology enables using AI for improved cancer diagnostics and ultimately better treatment decisions. We intend to develop a clinical-grade prognostic AI system for prostate cancer pathology by integrating: 1) Large-scale datasets of digitized prostate biopsies from a network of health care providers, 2) Cohorts with follow-up for clinical outcomes, 3) Genomic profiling of the samples, 4) Methods for calibration of scanner instruments and 5) Methods for estimating the reliability of predictions.
This will allow extending prognostication beyond human capacity by directly predicting disease progress from images and by integrating molecular data with tissue morphology. Utilizing diverse international data and calibration methods will allow training robust models and evaluating them in view of real-world sources of variation. Moreover, estimating the reliability of predictions is key for handling outliers and artefacts which are unavoidable in real-world use of diagnostic AI, and for providing useful feedback to the medical experts using the software in the clinic.
We will utilize the capacity of Berzelius AI/ML cluster by posing the project as an automated machine learning problem to design the system in a data-driven manner. This novel way of solving problems in computational pathology complements the major clinical significance of the project with the potential for technological advances. Once the system design and training is complete, we plan on validating the resulting AI solution in a real-world environment by conducting a clinical pilot study (2023), followed by a clinical trial (2024).