Pipelines for Automated Cell Recognition in TIFF and DICOM Images

Sammanfattning

This project was carried out in collaboration with Enaiblers AB with the aim of enabling their digital pathology platform ODEN to process and import whole-slide images as well as standard clinical imaging data stored in TIFF and DICOM formats. As TIFF and DICOM are widely used image formats in clinical environments, this work represents an initial step toward deploying ODEN in hospital-based diagnostic workflows. Two preprocessing pipelines were developed in Python, one for TIFF and one for DICOM. Both pipelines divide large images into overlapping tiles of a fixed size, extract spatial metadata such as microns per pixel and upload the resulting tiles to ODEN via the Enaiblers API. The DICOM pipeline additionally performs automatic modality detection using the DICOM SOP Class UID, routing whole-slide microscopy images to the wsidicom library and standard radiology modalities such as CT, MRI and X-ray to pydicom. Both pipelines are configurable through command-line arguments and use multithreaded execution to handle the large file sizes involved. Both pipelines were validated on real imaging data including a 90,600 × 92,008 pixel Pap smear whole-slide TIFF image, a whole-slide DICOM image and a set of 226 cranial CT slices. In all cases, tiles were successfully uploaded to ODEN and verified visually within the platform. The results confirm that the developed pipelines are functional and provide a working foundation for future annotation workflows and machine learning-based cell recognition within ODEN.