CIL: Open-Source Optimisation for Imaging Inverse Problems
Inverse problems arise when the quantity of interest cannot be measured directly, but must be estimated from indirect, noisy, incomplete, or degraded data. In scientific imaging, this includes tomographic reconstruction for medical imaging and materials science, denoising, deblurring, and inpainting. Although these applications look different, they share a common mathematical structure: recover an unknown image or volume by balancing agreement with the measured data against prior knowledge. This makes optimisation a central tool for solving inverse problems, but also creates a software challenge. How do we translate mathematical formulations into working code? How do we test different regularisers, constraints, and solvers? How do we compare optimisation strategies fairly and make experiments reproducible across datasets and applications? This talk presents the Core Imaging Library (CIL), an open-source Python framework for optimisation-based workflows in inverse problems. CIL provides composable tools for combining data readers, operators, objective functions and algorithms into transparent computational pipelines. The talk will show how CIL helps users prototype new methods, compare algorithms, and build reproducible workflows for applications in medical imaging, materials science, and other data intensive imaging modalities.
In many imaging applications, the measured data is only an indirect or degraded view of the object of interest. Optimisation provides a way to recover a meaningful image, volume, or signal by combining agreement with the data with prior knowledge about the expected solution. These assumptions may encode smoothness, sparsity, non-negativity, physical constraints and model or learned based priors. CIL is designed to make this optimisation structure explicit and user friendly . Instead of providing only fixed reconstruction routines, it offers reusable building blocks: data containers, operators, objective functions, regularisers and iterative algorithms. This allows users to move from a mathematical problem statement to executable code while keeping the workflow modular and inspectable.
In this talk, we will show how CIL supports optimisation-driven imaging workflows, from developing new methods to applying them to real experimental and industrial data. The goal is to demonstrate how open-source Python software can make advanced optimisation methods more usable, reproducible, and transferable across imaging applications:
- Academic research: advancing optimisation-driven imaging research through new algorithms, models, and reproducible experiments.
- Synchrotron facilities: supporting workflows for complex, high-dimensional experimental data.
- Industry: adapting imaging methods to real datasets and application-specific requirements.
We propose the following outline:
5 minutes: Imaging inverse problems and why they matter.
5 minutes: Overview of CIL as an open-source Python framework.
7 minutes: From maths to code: operators, functions and algorithms.
7 minutes: CIL in practice: tomography workflows for real data in medical imaging and materials science.
4 minutes: Reproducibility, benchmarking, and fair algorithm comparison.
2 minutes: Conclusions, future directions, and community contributions.
Evangelos Papoutsellis holds a PhD in Mathematical Image Processing from the University of Cambridge. His background spans academic research in France and the UK, industrial R&D, and open-source scientific software development. His work focuses on mathematical optimisation for inverse problems and computational imaging, with applications in medical imaging and materials science.