Project

Standardized Pipelines for the Accelerated Transfer of Advanced Imaging Software to Clinics

Short Summary

Most learning-based algorithms developed for clinical imaging are too demanding in terms of technical expertise to be deployed routinely in clinics. Hence, the development of standard approaches to these advanced algorithms is a foundational work of obvious importance. It is also an essential first step towards the integration of patients’ images with other “-omics” data. In this project, we shall design pipelines that simplify the access to automated digital imaging tools, hence helping to further spread their use in clinical settings. In parallel, we shall carry out two concrete research projects in collaboration with several clinical partners in Western Switzerland. We shall consider applications at both the microscopic scale (digital pathology) and the macroscopic scale (MRI), hence addressing two different yet complementary clinical needs.

Goals

In our quest for an accelerated deployment of imaging software technology in clinics, the goal of this project is twofold. First, we shall design standardized pipelines that allow clinicians to routinely, intuitively, and robustly use automated algorithms for the analysis of their patients’ images. Second, we shall address two concrete research questions of clinical relevance – in digital pathology and MRI brain imaging – for which new advanced ML-based image-analysis algorithms are needed.

Significance

The project addresses pressing needs for the standardized pipelines that facilitate the transfer and usage of novel image-analysis tools in clinical settings. By increasing the speed at which imaging software technology is brought to the Swiss clinical world, these pipe- lines should ultimately catalyze the research on personalized medicine and facilitate the eve- ryday analysis of patients’ data. In addition, innovative advanced image-analysis methods that address open questions in two major fields of clinical research—digital pathology and MRI brain imaging—will be developed throughout the project. Finally, all the algorithms developed throughout the project (and more) will be made available to the clinical community at large through the DeepImageJ plugin, hence enriching the practioners’ toolkit of automated methods.

Background

Clinical imaging is undergoing a revolution with the apparition of new learning-based algorithms that significantly improve the quality of reconstruction and facilitate the quantitative analysis of patient images. Yet, most machine-learning (ML) tools are developed using dedicated software frameworks (e.g., TensorFlow, PyTorch) that are still far too complex to handle for the medical community at large. This technical barrier hinders the transfer of advanced imaging software to clinicians, potentially adversely impacting their ability to fully exploit patients’ data and diagnose pathologies.

Platform

Consortium

Status

In Progress

4th Call, Platform