Development of surface correction software for medical image segmentation

Abstract

ARDS (Acute respiratory distress syndrome) is a severe and life-threatening respiratory syn- drome, characterized by lung inflammation, infiltration, alveolar edema and progressive hypox- emia. It is not the result of a specific condition, but rather a complication that can arise from existing conditions or as a result of infection and injuries. In most cases, patients require exten- sive care and mechanical breathing until the underlying condition is treated and the patient can use their lungs correctly. An essential step in ARDS treatment is the segmentation of CT(computerized tomography) images used to assess the lung conditions. However, such activity can be very time consuming if done manually, therefore, there is a lot of work being done to develop AI(Artificial Intelligence) automated methods that can provide rapid and accurate segmentations. Although the use of AI Has shown promising accuracy, errors can present themselves in some cases. Consequently, the development of such methods requires a mechanism for correcting erroneous segmentations, so that specialists can use them immediately on patients and to facilitate retraining of the existing AI models. Subsequently, this work focuses on the development and improvement of correction tools for resulting lung segmentations created by AI. The tools take advantage of the 3D mask created from the AI, utilize it as a 3D surface and operate directly on the mesh representing it, something which is not usually done on segmentation editing software. This way, the developed tools integrate numerous mesh editing techniques focused on the modification of a 3D mesh representing a lung surface. The development was done on the existing creaSDRA medical software which integrates multiple modules used in the segmentation process. Ultimately, the majority of the existing tools underwent substantial modifications to align with the specified requirements, augment their functionality, and optimize their usability. This comprehensive set of tools encompassed a deformation tool, a smoothing tool, and two surface- based tools. Moreover, in order to enhance the overall functionality and user experience of the software, two supplementary tools were developed, accompanied by multiple user interface (UI) changes. These enhancements collectively aimed to improve the efficacy and usability of the software solution.