Artificial intelligence in medical physics

Regular enrollment.

Specific use of artificial intelligence (AI) in medicine:

Use of convolutional neural networks (CNN) or similar in quantitative analysis of medical images
Use of recursive or similar networks in analysis of longitudinal data
AI in predictive models, comparsion to standard statistical methods

Methods of evaluation of appropriatnes of UI in medical applications:

Out of domain detection
Prediction uncertainty
Quantification of interpretability
Transfer and supplementary learning using new cases

LeCun, Y., Bengio, Y. & Hinton, G. Deep learning. Nature 521, 436–444 (2015)

Gonzalez C, Gotkowski K, Fuchs M, et al. Distance-based detection of out-of-distribution silent failures for Covid-19 lung lesion segmentation. Medical Image Analysis 82 2022: 102596.

Abdar, M, Pourpanah F, Hussain S, et al. A Review of Uncertainty Quantification in Deep Learning: Techniques, Applications and Challenges. Information Fusion 76 2021: 243-97.

Samek, W.; Binder, A.; Montavon, G.;Lapuschkin, S.; and M¨uller, K. 2017. Evaluating the visu-alization of what a deep neural network has learned. IEEE Trans. Neural Netw. Learning Syst. 28(11):2660–2673.

Pan, S.J., & Yang, Q. (2010). A Survey on Transfer Learning. IEEE Transactions on Knowledge and Data Engineering, 22, 1345-1359.

Students become familiar with general aspects of use of AI in medicine, and evaluation of performance for designated tasks in particular.

Course specific competences:

Understanding and identification of common principles of using AI in medicine. Familiarity with most common use cases. Understanding of methods of evaluation of AI for a particular purpuse. Familiarity with most common problems and limitations of AI in medicine.

Knowledge and understanding:

Physics and mathematics background of most common AI methods in medicine. Methods to evaluate fitness of AI to solve specific problems.

Use:

Examples of use of AI in medicine – prediction of risk in breast screening, segmentation of cancer on PET/CT images, identification of dementia based on PET/CT images. Practical evaluation of method uncertainty, out of distribution detection, interpretability quantification and transfer learning.

Reflection:

Understanding and identification of AI limitations. Illustration of performance drop for complex problems and decisions.

Lectures, homeworks, consultations.

Practical homework

Andrej Studen:

• Hribernik N, Huff DT, Studen A, et al. Quantitative imaging biomarkers of immune-related adverse events in immune-checkpoint blockade-treated metastatic melanoma patients: a pilot study. Eur J Nucl Med Mol Imaging. 2022;49(6):1857-1869.
• Klanecek Z, Wagner T, Wang YK et al. Uncertainty estimation for deep learning-based pectoral muscle segmentation via Monte Carlo dropout. Phys. Med. Biol. 68 2023: 115007.
• Studen A, Clinthorne N. System resolution versus image uncertainty for positron emission tomography scanners. J Med Imaging (Bellingham). 2022;9(3):033501.