A strategic and collaborative solution for systemic large-scale medical image data annotation

Purpose and goal:

We largely fulfill the first research goal. We first ported our interactive segmentation software MiaLab to web applications via web-assembly or and integrate them with the OHIF platform.

We also fulfilled the second research goal, where we integrate a deep learning-based lung lobe segmentation with the interactive image segmentation tool for strategic lung cancer sample prompt and iterative model + label refinement. Expected results and effects:

As a result of the proposed project, we developed a web-based medical image annotation tool that doesn´t need any installation and configuration on the user´s computer. It provides three different ways to perform 3D medical image annotation: region-based brush tool, mesh-based sculpting tool and levelset-based semi-automated smart painter. It is integrated with the widely used medical image viewer OHIF.

We believe the new tool could facilitate a large number of medical image AI projects that need image annotation from medical experts. Approach and implementation:

The benefit of porting MiaLab to a WebApp is that the user does not need installation. Moreover, it is no longer bound to a specific operating system. However, a downside is that the performance is worse than the native applications.

In our preliminary experiments, on modern computers, the three main annotation tools are still usable. Another downside is that the current Qt platform doesn´t support 3D rendering in a web-assembly application. Since most of the annotation tasks are done in 2D views than 3D views, the annotation function is still sufficient for most future projects.