Seminars 22/23
The TIA seminar is a regular event, organised by the Tissue Image Analytics Centre, that usually takes place on the 1st and 3rd Mondays of each month between 2pm and 3pm. We invite researchers and leaders in the area of computational pathology and associated areas to discuss their work and stimulate thought-provoking discussions.
Take a look at the details of TIA seminars during 2021-22 academic year, some of which are also available on our YouTube channel.
Please contact the TIA Seminar Organiser Dr Adam Shephard via email (adam.shephard@warwick.ac.uk) if you are interested in attending any of the seminars below.
A special thanks to Dr Simon Graham for setting up and organising the 21/22 series (see here).
Summer 2023
| Date | Speaker | Title |
|
15th May 2023 |
Azra Raza Columbia University Medical Center |
THE FIRST CELL and the human costs of pursuing cancer to the last |
Spring 2023
| Date | Speaker | Title |
|
16th January 2023 |
André Homeyer Fraunhofer MEVIS |
Artificial Intelligence in Pathology: From Prototype to Product
|
|
6th February 2023 |
Sarah Fremond Leiden University Medical Center |
|
|
20th February 2023 |
Oliver Lester Saldanha Uniklinik RWTH Aachen |
Swarm learning for decentralized deep learning in cancer histopathology
|
|
6th March 2023 |
Marc Aubreville Technische Hochschule Ingolstadt |
Mitosis Detection in Arbitrary Tumor Tissue - is the problem solved?
|
|
20th March 2023 |
Hanya Mahmood University of Sheffield |
|
|
18th April 2023 |
Paul Tourniaire Université Côte d'Azur |
Autumn 2022
| Date | Speaker | Title |
|
7th November 2022 |
Richard Chen Harvard Medical School |
Context-Aware and Multimodal Computational Pathology
|
|
21st November 2022 |
Mohamed Amgad Tageldin Northwestern University |
|
|
6th December 2022 |
James Diao Harvard Medical School |
|
|
19th December 2022 |
Korsuk Sirinukunwattana University of Oxford |
Image-based consensus molecular subtypes of colorectal cancer
|
Autumn 2022
Richard Chen
Harvard Medical School, Boston, USA (click here for a short bio)
7 Nov 2022
Title: Context-Aware and Multimodal Computational Pathology
Abstract: Survival outcome prediction is a challenging weakly-supervised and ordinal regression task in computational pathology that involves modeling complex interactions within the tumor microenvironment in gigapixel whole slide images (WSIs). Despite recent progress in formulating WSIs as bags for multiple instance learning (MIL), representation learning of entire WSIs remains an open and challenging problem, especially in overcoming: 1) the computational complexity of feature aggregation in large bags, and 2) the data heterogeneity gap in incorporating biological priors such as genomic measurements. In this talk, we present recent advances made in slide-level cancer prognostication using Transformer attention for early-based multimodal fusion, context modeling, interpretability. By equipping MIL frameworks with Transformer attention in weakly-supervised learning, we demonstrate improvements in AUC performance and localization of important morphological visual concepts, which may further aid the discovery of novel biomarkers.
Watch the seminar by clicking here.
Dr Mohamed Amgad Tageldin
Northwestern University, Chicago, USA (click here for a short bio)
21 Nov 2022
Title: Utilizing crowdsourcing approaches to curate large-scale datasets for computational pathology applications
Abstract: Deep learning enables accurate high-resolution mapping of cells and tissue structures that can serve as the foundation of interpretable machine-learning models for computational pathology. However, generating adequate labels for these structures is a critical barrier, given the time and effort required from pathologists. We describe a novel collaborative framework for engaging crowds of medical students and pathologists to produce quality labels for cell nuclei. We used this approach to produce the NuCLS dataset, containing >220,000 annotations of cell nuclei in breast cancers. This builds on prior work labeling tissue regions to produce the BCSS dataset to enable the multi-scale identification of tissue regions and nuclei within the breast cancer microenvironment. We present analysis results for single and multi-rater annotations from both non-experts and pathologists. We also share a novel workflow that uses algorithmic suggestions to collect accurate segmentation data without the need for laborious manual tracing of nuclei. Our results indicate that even noisy algorithmic suggestions do not adversely affect pathologist accuracy and can help non-experts improve annotation quality. Finally, we present a new approach for inferring truth from multiple raters and show that non-experts can produce accurate annotations for visually distinctive classes.
Watch the seminar by clicking here.
James A Diao
Harvard Medical School, Boston, USA (click here for a short bio)
6 Dec 2022
Title: Human-interpretable image features derived from densely mapped cancer pathology slides predict diverse molecular phenotype
Abstract: Computational methods have made substantial progress in improving the accuracy and throughput of pathology workflows for diagnostic, prognostic, and genomic prediction. Still, lack of interpretability remains a significant barrier to clinical integration. We present an approach for predicting clinically-relevant molecular phenotypes from whole-slide histopathology images using human-interpretable image features (HIFs). Our method leverages >1.6 million annotations from board-certified pathologists across >5700 samples to train deep learning models for cell and tissue classification that can exhaustively map whole-slide images at two and four micron-resolution. Cell- and tissue-type model outputs are combined into 607 HIFs that quantify specific and biologically-relevant characteristics across five cancer types. We demonstrate that these HIFs correlate with well-known markers of the tumor microenvironment and can predict diverse molecular signatures, including expression of four immune checkpoint proteins and homologous recombination deficiency, with performance comparable to ‘black-box’ methods. Our approach provides a comprehensive, quantitative, and interpretable window into the composition and spatial architecture of the tumor microenvironment.
Watch the seminar by clicking here.
Dr Korsuk Sirinukunwattana
University of Oxford, Oxford, UK (click here for a short bio)
19 Dec 2022
Title: Image-based consensus molecular subtypes of colorectal cancer
Abstract: In colorectal cancer (CRC), conventional histological grading is of little clinical value as it does not provide an insight into genetic/molecular events that drive the disease outcome. Image analysis of tumour tissue slides generated during a diagnostic examination of patients is a cost-effective tool to associate complex features of tissue organisation of CRC tumours with molecular and outcome data. Here we predict consensus molecular subtypes (CMS) of CRC from digital whole-slide images of CRC tissue sections using an ensemble of deep convolutional neural networks (Inception V3). Domain adversarial training and validation of the neural networks were performed using 1,553 tissue sections with comprehensive multi-omic data from three independent datasets (FOCUS trial, n=362 stage IV CRC; rectal cancer biopsies, n=175; The Cancer Genome Atlas (TCGA), n=572). Image-based consensus molecular subtyping (imCMS) accurately classified CRC whole-slide images and preoperative biopsies and spatially resolved intratumoural heterogeneity. In all three cohorts, imCMS reproduced expected correlations with (epi)genomic alterations and effectively stratified patients into prognostic subgroups. Leveraging artificial intelligence for the development of novel biomarkers extracted from histological slides with molecular and biological interpretability has remarkable potential for clinical translation.
Watch the seminar by clicking here.
Spring 2023
Dr André Homeyer
Fraunhofer MEVIS, Bremen, Germany (click here for a short bio)
16 Jan 2023
Title: Artificial Intelligence in Pathology: From Prototype to Product
Abstract: Despite thousands of published research papers on applications of artificial intelligence (AI) in pathology, only few prototypes have matured into commercial products for routine use. This talk will provide a brief overview and advice on various challenges that must be overcome before an AI solution can be brought to market. Topics covered include integration into the laboratory IT infrastructure, business models and reimbursement options. The talk will also address how to obtain regulatory approval for in vitro diagnostic medical devices and how to compile appropriate test datasets to evaluate real-world performance.
Watch the seminar by clicking here.
Sarah Fremond
Leiden University Medical Center, Leiden, Netherlands (click here for a short bio)
6 Feb 2023
Title: Interpretable deep learning model to predict the molecular classification of endometrial cancer from H&E-stained whole-slide images: a combined analysis of the PORTEC randomised trials and clinical cohorts
Abstract: Endometrial cancer can be molecularly classified into POLEmut, mismatch repair deficient (MMRd), p53 abnormal (p53abn), and no specific molecular profile (NSMP) subgroups. We aimed to develop an interpretable deep learning pipeline for whole-slide-image-based prediction of the four molecular classes in endometrial cancer (im4MEC), to identify morpho-molecular correlates, and to refine prognostication. im4MEC was trained to learn tile-level morphological features with self-supervised learning and to molecularly classify whole-slide images with an attention mechanism. Predictions of a nuclear classification deep learning model serve to derive interpretable morphological features. im4MEC attained macro-average AUROCs of 0,874 (95% CI 0,856–0,893) on four-fold cross-validation and 0,876 on the independent test set, PORTEC-3 a randomized trial of high risk patients. 5-year recurrence-free survival by im4MEC predicted molecular classes significantly stratified patients from PORTEC-3 (log-rank p<0,0001). POLEmut and MMRd tiles had a high density of lymphocytes, p53abn tiles had strong nuclear atypia, and NSMP had a low tumour-to-stroma ratio, a potentially novel characteristic feature.
Oliver Lester Saldanha
Uniklinik RWTH Aachen, Aachen, Germany (click here for a short bio)
20 Feb 2023
Title: Swarm learning for decentralized deep learning in cancer histopathology
Abstract: A limitation in computational pathology is the difficulty of exchanging data. Swarm Learning (SL) is a decentralized protocol for training deep learning models. In this study, we evaluated the effectiveness of SL in predicting microsatellite instability (MSI) and BRAF mutation status from histopathology images of colorectal and gastric cancers. We successfully used SL on large, multi-centric datasets of gigapixel histopathology images from more than 5,000 patients. Our results demonstrate that AI models trained using SL can accurately predict the BRAF mutational status and microsatellite instability directly from Hematoxylin and Eosin (H&E) stained pathology slides of colorectal and gastric cancers. It is crucial to have large datasets in the computational pathology of cancer, ideally from multiple centres, to minimize bias. However, the collection of such datasets faces practical, ethical, and legal obstacles. SL provides a potential solution to overcome these obstacles and may serve as an alternative for sharing patient-related data across various locations in the future.
Watch the seminar by clicking here.
Prof Marc Aubreville
Technische Hochschule Ingolstadt, Ingolstadt, Germany (click here for a short bio)
6 Mar 2023
Title: Mitosis Detection in Arbitrary Tumor Tissue - is the problem solved?
Abstract: The frequency of cells dividing, known as mitotic count, is strongly related to tumor growth and survival rates. However, assessing mitotic count is subjective and challenging for human experts, particularly identifying the most prognostic area with the highest mitotic count. Consequently, automatic detection of mitotic figures using computer vision has gained attention. The Mitosis Domain Generalization (MIDOG) challenges evaluate the quality of methods for detection in datasets that may differ significantly in appearance due to variations between tumor subtypes or imaging centers. Several domain generalization methods have been tested in these challenges, with many surpassing human performance. Nevertheless, despite achieving superhuman performance, we still need to identify and explore factors that may limit robustness in real-world scenarios. Therefore, this talk aims to investigate unexplored factors that may affect detection accuracy, even with advanced methods in place.
Watch the seminar by clicking here.
Dr Hanya Mahmood
University of Sheffield, Sheffield, UK (click here for a short bio)
20 March 2023
Title: Prediction of malignant transformation and recurrence of oral epithelial dysplasia using architectural and cytological feature specific prognostic models
Abstract: Oral epithelial dysplasia (OED) is a precursor state usually preceding oral squamous cell carcinoma. Histological grading is the gold standard for OED prognostication but is subjective with inconsistent outcome prediction. This talk will present the findings from a recently published study, which explores if individual OED histological features can be used to develop predictive models for malignant transformation and recurrence. Two models will be discussed: a ‘6-point model’ using the six features showing greatest prognostic association and a ‘two-point model’ using the two features with highest inter-rater pathologist agreement. Both models showed good predictive ability for both transformation and recurrence, with further improvements when certain clinical variables were added. The findings demonstrate a correlation between individual OED histological features and prognosis for the first time. The proposed models have the potential to simplify OED grading and improve patient management.
Watch the seminar by clicking here.
Paul Tourniaire
Université Côte d'Azur, Nice, France
18 April 2023
Title: TBC
Summer 2023
Prof Azra Raza
Columbia University Medical Center, New York, USA (click here for a short bio)
3 April 2023
This is a special edition of our TIA Centre Seminar Series, held in collaboration with the TIA Centre, Health GRP and Warwick Cancer Centre
Title: THE FIRST CELL and the human costs of pursuing cancer to the last
Abstract: Cancer will strike 1 in 2 men and 1 in 3 women. Yet, we are failing spectacularly to improve outcome for the majority of patients. Our contention is that the real solution to the cancer problem is to diagnose cancer early, at the stage of The First Cell. The footprints of early cancers, the surrogate biomarkers, need to be identified so we can detect the first cell. The rapidly evolving technologies are doing much in this area but need to be expanded. We study a pre-leukemic condition called myelodysplastic syndrome (MDS) with the hope that we can detect the first leukemia cells as the disease transforms to acute myeloid leukemia (AML). Towards this end, we have collected blood and bone marrow samples on MDS and AML patients since 1984. Today, our Tissue Repository has more than 60,000 samples. We propose novel methods to identify surrogate markers that can identify the First Cell through studying the serial samples of patients who evolve from MDS to AML.