Computational and Numerical Methods for Medical Physics

4.00 credits

24.0 h + 10.0 h

Teacher(s)

. SOMEBODY; Lee John; Sterpin Edmond;

Language

English

Main themes

This course aims to familiarize the student with the computational and numerical methods frequently used in Medical Physics, like Monte Carlo simulations, with their physical and statistical underlying concepts, as well as to provide the basis of Artificial Intelligence, Machine Learning, and Deep Learning techniques and their use to solve data-driven problems.

Learning outcomes

At the end of this learning unit, the student is able to :
The specific learning outcomes are: The student is able to apply advanced statistical methods needed in Medical Physics. The student acquires the computational and calculation skill to address quantitatively common problems in Medical Physics. The student solves classification and regression problems in Medical Physics by applying Machine Learning and Artificial Intelligence techniques like decision trees, random forests, neural networks, deep learning, ... to various types of data, including medical images. The student understands and masters the basic aspects of optimization methods that underpin most of the aforementioned techniques.

Content

The course will be organized around three main pillars

Advanced Statistics in Medical Physics: statistics are heavily used in medicine in general and in medical physics in particular. This includes statistical significance of laboratory and clinical experiments; quantification of risk; estimation and propagation of uncertainties (type A and type B uncertainties); probabilistic problem solving.
Monte Carlo techniques. Monte Carlo engines are often used as black box in clinical practice and R&D. The goal is to provide insights in the theoretical grounds of Monte Carlo simulations and also in the practical specificities of modern implementations. This includes: random number generation; sampling techniques (inverse transform, rejection technique); variance reduction; statistical error estimation (direct or batch technique); problem definition (geometry and materials); use of specialized hardware (many-core processors and GPU). Practical examples and important results are illustrated in radiotherapy, nuclear medicine and radiology.
Introduction to Machine Learning:
- Context and purpose of Artificial Intelligence, Machine Learning, and Deep Learning.
- The various types of learning problems (supervised, non-supervised, reinforcement, transfer).
- The various types of data sets and their purpose (training, validation, test).
- Short introduction to optimization.
- A few techniques:
  - Principal component analysis, linear discriminant analysis.
  - Decision trees and random forests.
  - Support Vector Machines.
- Neural networks, from single artificial neuron to deep (convolutional) networks.oInterpretation of the results (ROC curve/sensitivity/specificity/...).
- Specifics of data collection for AI/ML/DL in medical physics (access to patient data and the importance of consistent patient data; how to guide efforts in structured reporting).
- Big data and data preprocessing (images, radiomics,.. ).

Teaching methods

Lectures
For each part of the course (statistics, Monte Carlo techniques, Machine Learning) a series of exercises is proposed. They should be solved by using the adequate computing material.

Evaluation methods

Type : Partial or continuous assessment with (final) exam during the examination period
Description of evaluation : Oral, Report

Exam during the examination period: 70% of the total score.
Reports on the assignments of selected exercises: 30% of the total score.

Faculty or entity

PHYS

Programmes / formations proposant cette unité d'enseignement (UE)

Title of the programme

Sigle

Credits

Prerequisites

Learning outcomes

Master [120] in Physics [professional focus of Medical Physics : UCLouvain-KULeuven]

PHYS2M