My team at Roche-genentech actively worked on computational pathology tools for scoring tumor infiltrating lymphocytes in H&E slides and identifying histology-derived features for stratifying patients in breast and lung cancers.

Accurate subtyping of ovarian carcinomas is an increasingly critical and often challenging diagnostic process. This work focuses on the development of an automatic classification model for ovarian carcinoma subtyping. Specifically, we present a novel clinically inspired contextual model for histopathology image subtyping of ovarian carcinomas. A whole slide image is modelled using a collection of tissue patches extracted at multiple magnifications. An efficient and effective feature learning strategy is used for feature representation of a tissue patch. The locations of salient, discriminative tissue regions are treated as latent variables allowing the model to explicitly ignore portions of the large tissue section that are unimportant for classification. These latent variables are considered in a structured formulation to model the contextual information represented from the multi-magnification analysis of tissues. A novel, structured latent support vector machine formulation is defined and used to combine information from multiple magnifications while simultaneously operating within the latent variable framework.

We focus on the problem of stain inconsistency in the context of automatic histopathology image analysis and we aim at creating a model that would ultimately facilitate the applicability of automatic histopathology image analysis systems, such as nuclei segmentation or cancer classification systems, across pathology labs and regardless of the image acquisition procedure. Specifically, we propose a novel methodology for transferring stains across different datasets describing a similar pathology but with different staining appearance. We assume that such a model should couple stain normalization with the image analysis task without involving pre-processing images or re-training the analysis system. Also, we believe such system should not rely on a single given template reference image but should rather learn how best to transfer stains across datasets by leveraging datasets distributions. Based on these assumptions, we propose a fully trainable framework in which stain normalization is modelled as an adversarial game and is performed jointly with a specific image analysis task.

We present a novel attention based model for predicting cancer from histopathology whole slide images. The proposed model is capable of attending to the most discriminative regions of an image by adaptively selecting a limited sequence of locations and only processing the selected areas of tissues. We demonstrate the utility of the proposed model on the slide-based prediction of macro and micro metastases in sentinel lymph nodes of breast cancer patients. We achieve competitive results with state-of-the-art convolutional networks while automatically identifying discriminative areas of tissues.