Whole Slide Images (WSI) are widely used in histopathology for research and the diagnosis of different types of cancer. The preparation and digitization of histological tissues leads to the introduction of artifacts and variations that need to be addressed before the tissues are analyzed. WSI preprocessing can significantly improve the performance of computational pathology systems and is often used to facilitate human or machine analysis. Color processing techniques are usually the main concern, while other areas are frequently ignored. In this paper, we present a detailed study of the state-of-the-art in three different areas of WSI preprocessing: Artifacts detection, color variation, and the emerging field of pathology-specific data augmentation. We include a summary of evaluation techniques along with a discussion of possible limitations and future research directions for new methods.

Histological tissue examination has been a longstanding practice for cancer diagnosis where pathologists identify the presence of tumors on glass slides. Slides acquired from laboratory routine may contain unintentional artifacts due to complications in surgical resection. Blood and damaged tissue artifacts are two common problems associated with transurethral resection of the bladder tumor. Differences in histotechnical procedures among laboratories may also result in color variations and minor inconsistencies in outcome. A digitized version of a glass slide known as a whole slide image (WSI) holds enormous potential for automated diagnostics. The presence of irrelevant areas in a WSI undermines diagnostic value for pathologists as well as computational pathology (CPATH) systems. Therefore, automatic detection and exclusion of diagnostically irrelevant areas may lead to more reliable predictions. In this paper, we are detecting blood and damaged tissue against diagnostically relevant tissue. We gauge the effectiveness of transfer learning against training from scratch. Best models give 0.99 and 0.89 F1 scores for blood and damaged tissue detection. Since blood and damaged tissue have subtle color differences, we assess the impact of color processing methods on the binary classification performance of five well-known architectures. Finally, we remove the color to understand its importance against morphology on classification performance.

No abstract

Bladder cancer patients' stratification into risk groups relies on grade, stage and clinical factors. For non-muscle invasive bladder cancer, T1 tumours that invade the subepithelial tissue are high-risk lesions with a high probability to progress into an aggressive muscle-invasive disease. Detecting invasive cancerous areas is the main factor for dictating the treatment strategy for the patient. However, defining invasion is often subject to intra/interobserver variability among pathologists, thus leading to over or undertreatment. Computer-aided diagnosis systems can help pathologists reduce overheads and erratic reproducibility. We propose a multi-scale model that detects invasive cancerous areas patterns across the whole slide image. The model extracts tiles of different tissue types at multiple magnification levels and processes them to predict invasive patterns based on local and regional information for accurate T1 staging. Our proposed method yields an F1 score of 71.9, in controlled settings 74.9, and without infiltration 90.0.

No abstract