Automatic Multi-Stain Registration of Whole Slide Images in Histopathology
Authors:
Abubakr Shafique,
Morteza Babaie,
Mahjabin Sajadi,
Adrian Batten,
Soma Skdar,
H. R. Tizhoosh
Abstract:
Joint analysis of multiple biomarker images and tissue morphology is important for disease diagnosis, treatment planning and drug development. It requires cross-staining comparison among Whole Slide Images (WSIs) of immuno-histochemical and hematoxylin and eosin (H&E) microscopic slides. However, automatic, and fast cross-staining alignment of enormous gigapixel WSIs at single-cell precision is ch…
▽ More
Joint analysis of multiple biomarker images and tissue morphology is important for disease diagnosis, treatment planning and drug development. It requires cross-staining comparison among Whole Slide Images (WSIs) of immuno-histochemical and hematoxylin and eosin (H&E) microscopic slides. However, automatic, and fast cross-staining alignment of enormous gigapixel WSIs at single-cell precision is challenging. In addition to morphological deformations introduced during slide preparation, there are large variations in cell appearance and tissue morphology across different staining. In this paper, we propose a two-step automatic feature-based cross-staining WSI alignment to assist localization of even tiny metastatic foci in the assessment of lymph node. Image pairs were aligned allowing for translation, rotation, and scaling. The registration was performed automatically by first detecting landmarks in both images, using the scale-invariant image transform (SIFT), followed by the fast sample consensus (FSC) protocol for finding point correspondences and finally aligned the images. The Registration results were evaluated using both visual and quantitative criteria using the Jaccard index. The average Jaccard similarity index of the results produced by the proposed system is 0.942 when compared with the manual registration.
△ Less
Submitted 29 July, 2021;
originally announced July 2021.
A Sample-Based Training Method for Distantly Supervised Relation Extraction with Pre-Trained Transformers
Authors:
Mehrdad Nasser,
Mohamad Bagher Sajadi,
Behrouz Minaei-Bidgoli
Abstract:
Multiple instance learning (MIL) has become the standard learning paradigm for distantly supervised relation extraction (DSRE). However, due to relation extraction being performed at bag level, MIL has significant hardware requirements for training when coupled with large sentence encoders such as deep transformer neural networks. In this paper, we propose a novel sampling method for DSRE that rel…
▽ More
Multiple instance learning (MIL) has become the standard learning paradigm for distantly supervised relation extraction (DSRE). However, due to relation extraction being performed at bag level, MIL has significant hardware requirements for training when coupled with large sentence encoders such as deep transformer neural networks. In this paper, we propose a novel sampling method for DSRE that relaxes these hardware requirements. In the proposed method, we limit the number of sentences in a batch by randomly sampling sentences from the bags in the batch. However, this comes at the cost of losing valid sentences from bags. To alleviate the issues caused by random sampling, we use an ensemble of trained models for prediction. We demonstrate the effectiveness of our approach by using our proposed learning setting to fine-tuning BERT on the widely NYT dataset. Our approach significantly outperforms previous state-of-the-art methods in terms of AUC and P@N metrics.
△ Less
Submitted 15 April, 2021;
originally announced April 2021.
