TIA Seminars
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 some of the previous seminars on our YouTube channel.
Please contact the TIA Seminar Organiser Dr Simon Graham via email (Simon dot Graham at warwick dot ac dot uk) if you are interested in attending any of the seminars below.
Summer 2022
| Date | Speaker | Title |
|
16th May 2022 |
Mattias Rantalainen
Karolinska Institute
|
|
|
6th June 2022 |
Jana Lipkova
Harvard Medical School
|
Deep learning-enabled assessment of cardiac allograft rejection from endomyocardial biopsies |
|
27th June 2022 |
Nick Trahearn The Institute of Cancer Research |
|
|
4th July 2022 |
Nikolas Stathonikos
UMC Utrecht
|
TBA |
|
18th July 2022 |
TBA
|
TBA |
Spring 2022
| Date | Speaker | Title |
|
17th January 2022 |
Wouter Bulten
Radboud University Medical Center
|
|
|
7th February 2022 (4pm) |
Robert Noble
City University of London
|
|
|
21st February 2022 |
Johannes Lotz
Fraunhofer Institute for Digital Medicine
|
High-Resolution Image Registration for Computational Pathology |
|
7th March 2022 |
Ke Yuan
University of Glasgow
|
|
|
28th March 2022 |
Jane Armes
The University of the Sunshine Coast
|
|
|
4th April 2022 |
Andrew Schaumberg
Harvard Medical School
|
|
|
19th April 2022 |
Chensu Xie
Memorial Sloan Kettering Cancer Center
|
Autumn 2021
| Date | Speaker | Title |
| 11th October 2021 |
Saad Nadeem Memorial Sloan Kettering Cancer Center |
Mathematical Oncology Initiative |
| 18th October 2021 |
Pushpak Pati & Guillaume Jaume IBM Research Zurich |
HistoCartography: Entity-graph representations and models in Computational Pathology |
| 1st November 2021 |
Maxime Lafarge University Hospital of Zurich |
Deep Learning and the Arbitrary Orientation of Histology Images |
| 15th November 2021 |
Faisal Mahmood Harvard Medical School |
Data-Efficient and Multimodal Computational Pathology |
| 6th December 2021 |
Narmin Ghaffari Laleh RWTH University Hostpital |
Benchmarking Artificial Intelligence Methods for End-to-End Computational Pathology |
| 20th December 2021 |
Çiğdem Gündüz Demir Koç University |
Segmentation Networks for Digital Pathology |
Autumn 2021
Dr Saad Nadeem
Memorial Sloan Kettering Cancer Center, NYC, USA (click here for a short bio)
11 Oct 2021
Title: Mathematical Oncology Initiative
Abstract: In this talk, I will present some of our recent work on addressing challenges in analysing H&E, IHC, and multiplex images. Specifically, I will talk about (1) advanced mathematical approaches (multi-marginal optimal mass transport) for stain normalisation and augmentation (as one way to achieve stain invariance), (2) advanced mathematical algorithms (vectorial optimal mass transport) for robust uni-/cross-modal (H&E/IHC/multiplex) WSI registration, (3) stain-invariant multitask deep learning framework, DeepLIIF, for simultaneous stain deconvolution/separation, cell segmentation, and quantitative single-cell IHC scoring (this framework is now deployed in Paige.ai's Slide Viewer), (4) interactive deep learning framework, ImPartial, for whole-cell segmentation and auto-thresholding in multiplex images, and (5) advanced mathematical (spectral analysis + optimal mass transport + Ricci curvature + non-abelian Fourier analysis) Spatial Profiling Toolbox for spatial biomarker quantification and discovery in digital pathology images. I will conclude with future directions and brief overview of our recent works in radiation oncology and surgery.
Watch the seminar by clicking here.
Pushpak Pati & Guillaume Jaume
IBM Research, Zurich, Switzerland (click here for short bios)
18 Oct 2021
Title: HistoCartography: Entity-graph representations and models in Computational Pathology
Abstract: Computational pathology aims to provide computer-aided diagnosis tools to automate, assist and augment pathologists. To this end, we introduce you to HistoCartography that leverages biological entities and Graph Neural Networks for the modelling and analysis of histology images. In the first part, we will present a novel hierarchical cell-to-tissue-graph (HACT) representation for comprehensively encoding tissue regions. We will further introduce HACT-Net, a neural network to map HACT to histopathological breast cancer subtypes. The power of HACT-Net will be demonstrated on BRACS, a novel dataset for BReAst Cancer Subtyping of histology RoIs. In the second part, we will emphasize on the need for explainability to assist medical decisions and build trust between pathologists and AI algorithms. We will see how entity-graph processing can enable the explainability in pathologist-friendly terms. Finally, we will emphasize the scalability aspect of entity-graphs by introducing SegGini. It aims to semantically segment gleason patterns in prostate TMAs and WSIs by leveraging weak image-level supervision.
Watch the seminar by clicking here.
Dr Maxime Lafarge
University Hospital of Zurich, Switzerland (click here for a short bio)
1 Nov 2021
Title: Deep Learning and the Arbitrary Orientation of Histology Images
Abstract: As deep learning has become the methodology of choice for image classification tasks, the robustness of deep learning models against irrelevant factors of variation is a necessary property for their safe deployment, especially in clinical environments. Solutions based on geometric deep learning have enabled models to gain invariance properties, thus bringing robustness guarantees against specific irrelevant variations. This talk will focus on the problem of the robustness to the arbitrary orientation of histology images in the context of supervised and unsupervised learning frameworks for computational pathology. Further, I will discuss the limitations of standard convolutional neural networks, how to expand existing architectures with rotation-equivariant operations and to what extent this is a solution.
Dr Faisal Mahmood
Harvard Medical School, Boston, USA (click here for a short bio)
15 Nov 2021
Title: Data-efficient and multimodal computational pathology
Abstract: Advances in digital pathology and artificial intelligence have presented the potential to build assistive tools for objective diagnosis, prognosis and therapeutic-response and resistance prediction. In this talk we will discuss: 1) Data-efficient methods for weakly-supervised whole slide classification with examples in cancer diagnosis and sub-typing (Nature BME, 2021), and allograft rejection (Nature Medicine, 2021) 2) Harnessing weakly-supervised, fast and data-efficient WSI classification for identifying origins for cancers of unknown primary (Nature, 2021). 3) Discovering integrative histology-genomic prognostic markers via interpretable multimodal deep learning (IEEE TMI, 2020; ICCV, 2021). 4) Federated learning for computational pathology (MedIA, 2021). 5) Deploying weakly supervised models in low resource settings without slide scanners, network connections, computational resources and expensive microscopes. 6) Bias and fairness in computational pathology algorithms.
Watch the seminar by clicking here.
Narmin Ghaffari Laleh
RWTH University Hospital, Aachen, Germany (click here for a short bio)
6 Dec 2021
Title: Benchmarking artificial intelligence methods for end-to-end computational pathology
Abstract: Artificial intelligence (AI) can extract subtle visual information from digitised histopathology slides and yield biological insight on genotype-phenotype interactions as well as clinically actionable recommendations. In the last years, diverse development of the computational techniques resulted in advanced algorithms like Multiple-Instance learning (MIL) and clustering-constrained attention MIL (CLAM), which their performances have been evaluated on specific and clinically relevant targets. However, it is unclear how these different approaches perform relative to each other. We performed a large-scale benchmarking study and implemented and systematically compared five methods in six clinically relevant end-to-end prediction tasks. In particular, we compared a classical residual neural network (ResNet), a modern convolutional neural network (EfficientNet), and non-convolutional vision transformers (ViT) with classical MIL and CLAM. We showed that histological tumour sub-typing of renal cell carcinoma is an easy task which all approaches can successfully solve with an area under the receiver operating curve (AUROC) of above 0.9 without any significant differences between approaches. In contrast, we report significant performance differences for mutation prediction in colorectal, gastric, and bladder cancer. Weakly supervised ResNet- and ViT-based workflows significantly outperformed other methods, in particular MIL and CLAM for mutation prediction. This benchmark study provides important actionable advice for future studies and real-world applications of computational pathology.
Watch the seminar by clicking here.
Dr Çiğdem Gündüz Demir
Koç University, Istanbul, Turkey (click here for a short bio)
20 Dec 2021
Title: Segmentation Networks for Digital Pathology
Abstract: Digital pathology emerges as a new field thanks to the recent technological developments in high-throughput slide scanners, which enable substantially massive imaging capacity, as well as the progressively growing processor, storage, and network capabilities. An important complementary part of this new field is to make robust computational image analysis tools available. Segmenting regions of interest on a digital pathology slide is typically the first but one of the foremost steps of these tools, which greatly affects the success of the entire analysis. In this talk, I will briefly mention the main challenges associated with segmentation tasks in digital pathology and microscopic image analysis, and then present examples of the dense prediction networks that my research group designed and implemented to address these challenges. Particularly, I will talk about our proposed network architectures and loss functions that were specifically designed to facilitate better training of the segmentation networks. At the end, I will discuss future research possibilities towards the direction of developing more robust segmentation networks for digital pathology.
Watch the seminar by clicking here.
Spring 2022
Dr Wouter Bulten
Radboud University Medical Center, The Netherlands (click here for a short bio)
17 Jan 2022
Title: Artificial intelligence as a digital fellow in pathology: Human-machine synergy for improved prostate cancer diagnosis
Abstract: For patients with prostate cancer, the Gleason grading system used by pathologists is a crucial part of the diagnosis and significantly influences treatment planning. However, while the grading system is the strongest tissue-based marker, it is inherently subjective. Consequently, it exhibits high variability between pathologists and even within the same pathologist at different points in time.
In this talk, I present an overview of my Ph.D. research in which we set out to improve prostate cancer diagnostics through deep learning. I will show that AI algorithms can achieve pathologist-level performance at highly complex tasks like Gleason grading.
I will also present the first results of the PANDA Challenge: an international competition on AI for Gleason grading in biopsies. The challenge, organized by Radboud University Medical Center, Karolinska Institute, and Google Health, was the biggest challenge in pathology to date with 10k+ biopsies for training, 2k+ biopsies for independent validation, and over a thousand participants.
Finally, I will demonstrate that AI can successfully assist pathologists; together, they can improve the diagnostic process to ultimately deliver a more accurate diagnosis to patients.
Watch the seminar by clicking here.
Dr Robert Noble
City University of London, United Kingdom (click here for a short bio)
7 Feb 2022 (4pm)
Title: Explaining the Modes of Tumour Evolution
Abstract: Understanding the nature of tumour evolution underpins accurate prognosis and the design of effective treatment strategies. Whereas selective sweeps are prevalent during early tumour growth, later stages exhibit either sparse branching or effectively neutral evolution. I will present new insights into the causes and consequences of these different patterns based on mathematical analysis, computational models, and analysis of clinical data. I will show that, within biologically relevant parameter ranges, different spatial structures can generate distinct tumour evolutionary modes. These model predictions are moreover consistent with data for cancer types with corresponding spatial structures. Surprisingly simple mathematical expressions can be derived to explain why selective sweeps are rare except when tumours are relatively very small. I will further introduce indices for categorising evolutionary modes, including a new class of robust, universal tree balance indices. Although my work is motivated by questions in cancer research, many of these results are readily applicable to other systems.
Watch the seminar by clicking here.
Dr Johannes Lotz
Fraunhofer Institute for Digital Medicine MEVIS, Germany (click here for a short bio)
21 Feb 2022
Title: High-Resolution Image Registration for Computational Pathology
Abstract: Image registration in pathology is (mostly) used to correlate information from differently stained consecutive or re-stained sections. In this talk, I will give a brief introduction into the concepts behind image registration, highlight some applications and demonstrate how registration can be used in practice. An important distinction is made between re-stained and consecutive sections and we will see some preliminary results from our evaluation of image registration applied to these two techniques.
Watch the seminar by clicking here.
Dr Ke Yuan
University of Glasgow, United Kingdom (click here for a short bio)
7 Mar 2022
Title: Learning Deep Representations of Cancer Tissue
Abstract: Cancer is an evolutionary process characterised by heterogeneity between and within tumours. In this talk, I will discuss how unsupervised representation learning models can improve quantification of heterogeneity within histological images of tumour slides. These representation learning models include deep generative models and self-supervised models. Using images across Breast, Colon and Lung tumours, these models capture distinct phenotypic characteristics of tissue samples, including cancer cell destiny, tissue types, growth patterns and clinical outcome, paving the way for further understanding of tumour progression and tumour micro-environment, and ultimately refining histopathological classification for diagnosis and treatment.
Watch the seminar by clicking here.
Dr Jane Armes
University of the Sunshine Coast, Queensland, Australia (click here for a short bio)
28 Mar 2022
Title: Computational Pathology: Diverse Applications – From Triple Negative Breast Cancer to Perinatal Death
Abstract: Jane Armes is a histopathologist of more than 30 years’ diagnostic experience. In this talk she discusses two areas in which diagnostic pathology would greatly benefit from integration with Computational Pathology. Using the example of Triple Negative Breast Cancer, she illustrates the importance of a cancer’s morphology as the manifestation of its specific genetic drivers interacting with its specific host response and how Computational Pathology could be used to quantitatively harness prognostic and predictive data not otherwise extractable from histology sections. She then discusses the complex, multi-modal investigations necessary to interrogate the causes of Perinatal Death and how integration of literature-based, phenotypic, genetic and morphologic databases are necessary to leverage the investigation of perinatal death beyond standard autopsy-based documentation.
Watch the seminar by clicking here.
Dr Andrew Schaumberg
Harvard Medical School, Boston, USA (click here for a short bio)
4 April 2022
Title: A Computational Path from Microscope to Social Media
Abstract: What is important in a slide to a pathologist, and what may machines learn from that?
First, I will introduce a diagnostic saliency pilot study (PMID:29601065). Using a 3d-printed mount, we video-recorded pathologists making a diagnosis at a microscope. Video frames were registered to whole slide images to aggregate visual dwell time. We let saliency be a function of dwell time. We found a convolutional neural network could accurately distinguish short-dwell-time regions from long-dwell-time regions in these whole slide images. This suggested our methods were a tractable noninvasive way to collect pixel-level annotations for downstream supervised learning tasks.
Second, we noticed pathologists worldwide share on social media photomicrographs of anonymized patient cases (PMID:32467650). We collected photomicrographs of consenting pathologists and deployed a real-time patient case similarity search tool, which we named “Pathobot” on Twitter. Beyond search, for any hematoxylin-and-eosin-stained photomicrograph our methods accurately predicted disease state as non-neoplastic, benign/low-grade-malignant-potential, or malignant. To our knowledge this is the first study of photomicrographs shared on social media.
If time permits, we will briefly discuss our 3d-printed “Pathobox” project to mount a smartphone to a microscope and engage more pathologists, particularly those in low-resource settings.
https://thepathologist.com/diagnostics/pathobot-deep-learning-for-humans-and-machines
Watch the seminar by clicking here.
Dr Chensu Xie
Memorial Sloan Kettering Cancer Center, USA (click here for a short bio)
19 April 2022
Title: Computational biomarker predicts lung ICI response via deep learning-driven hierarchical spatial modelling from H&E
Abstract: Determining which lung cancer patients are likely to respond to immune checkpoint inhibitors (ICI) remains a crucial challenge. Existing FDA-approved biomarkers lack sensitivity and specificity for identifying treatment candidates. To overcome this problem, we present a computational biomarker for predicting ICI response directly from routine H&E stained whole slide images of the initial biopsy. To achieve this, we developed an end-to-end deep learning system (EPL-GNN) that performs hierarchical spatial modeling on whole slide images to learn both spatial and morphological features from 2.1 billion cells and output a response score for each patient. The computational biomarker was trained and evaluated on the largest reported cohort of stage 4 lung cancer patients with ICI treatment response (N=583), resulting in an AUC of 0.69 and sensitivity of 91% on the independent test cohort, which compares favorably to PD-L1 immunohistochemistry (IHC) with an AUC of 0.68 and sensitivity of 57%, and tumor mutation burden (TMB) with an AUC of 0.62. The EPL-GNN model correctly identified 81% of the responders with a negative PD-L1 IHC result. Visualizations of the hierarchical spatial model revealed potential cellular patterns that correspond to ICI treatment response. In addition to the increased sensitivity achieved by the EPL-GNN model, H&E-based Computational Biomarkers offer a faster, less expensive, more objective and reproducible alternative or adjunct to existing IHC or sequencing based biomarkers.
Watch the seminar by clicking here.
Dr Mattias Rantalainen
Karolinska Institute, Sweden (click here for a short bio)
16 May 2022
Title: AI-based computational pathology: enabling scalable and comprehensive phenotyping for patient stratification in cancer precision medicine
Abstract: Precision medicine has the potential to substantially improve cancer patient outcomes. However, to be effective, precision diagnostic solutions for patient stratification are required. To improve outcomes for broad groups of patients, fast, reliable and cost-effective solutions are needed. Current routine diagnosis of cancer is based on manual histopathological assessment, which is imprecise. Molecular diagnostics on the other hand offers improved patient stratification, but at a high price, limiting patient access and imposing a high economic burden on healthcare systems.
We develop, validate, translate and implement AI-based histopathology image analysis solutions (computational pathology) for image-based phenotyping and patient stratification, both in the clinical setting and for cancer research. Clinical pathology is currently undergoing a digital transition, which facilitates implementation of AI-based decision support tools for precision diagnostics that are fast and only cost a fraction compared with molecular diagnostics. Our research is based on applications of deep learning -based methods to model large population representative studies with gigapixel histopathology images, registry-based clinical information, and molecular profiling data.
In this seminar I will provide an overview of some of our on-going research, recent results and translational activities in the area of breast- and prostate cancer.
Watch the seminar by clicking here.
Dr Jana Lipkova
Harvard Medical School, USA (click here for a short bio)
6 June 2022
Title: Deep learning-enabled assessment of cardiac allograft rejection from endomyocardial biopsies
Abstract: Endomyocardial biopsy (EMB) screening represents the standard of care for detecting allograft rejections after heart transplant. Manual interpretation of EMBs is affected by substantial inter-observer and intra-observer variability, which often leads to inappropriate treatment with immunosuppressive drugs, unnecessary follow-up biopsies and poor transplant outcomes. Here we present a deep learning-based artificial intelligence (AI) system for automated assessment of gigapixel whole-slide images obtained from EMBs, which simultaneously addresses detection, subtyping and grading of allograft rejection. To assess model performance, we curated a large dataset from the United States, as well as independent test cohorts from Turkey and Switzerland, which includes large-scale variability across populations, sample preparations and slide scanning instrumentation. In a human reader study, the AI system showed non-inferior performance to conventional assessment and reduced inter-observer variability and assessment time. This robust evaluation of cardiac allograft rejection paves the way for clinical trials to establish the efficacy of AI-assisted EMB assessment and its potential for improving heart transplant outcomes.
Dr Nick Trahearn
The Institute of Cancer Research, UK (click here for a short bio)
27 June 2022
Title: FORECAST: How tumour evolvability metrics can predict long term recurrence in locally-advanced prostate cancer
Abstract: Cancers follow an evolutionary process, obeying Darwinian laws to produce adaptations that ultimately lead to phenomena such as recurrent disease, drug resistance, and tumour metastasis. Consequently, the study of cancer evolution can provide a strong basis for predictive oncology and, perhaps eventually, personalised medicine. However, the predictive power of evolutionary metrics in cancer have seldom been tested, and thus there is a need for quantitative measurements in controlled clinical trials with long term follow-up information. This is particularly true in locally-advanced prostate cancer, which can recur more than a decade after diagnosis.
In prostate pathology, the Gleason grading system remains the gold standard as a strong predictor of long term patient outcome. However the current 2014 ISUP standard may not be sufficient to adequately stratify high-risk patients. Thus, there remains a need for predictors of outcome beyond those in current diagnostic practice, and for which measures of evolvability offer a potential solution.
In this study we assessed 1923 diagnostic slides from 250 high-risk prostate cancer patients, who took part in the prostate IMRT trial at The Royal Marsden Hospital, for which full clinical information and 12y median follow-up was available. H&E tissue sections from this cohort were analysed with deep learning models to segment glands and classify them by their associated Gleason pattern. The Gleason grading by deep learning was able to identify subgroups of patients at higher risk of recurrence, both through standard ISUP grade grouping and a novel "evolvability metric", based on heterogeneity of Gleason pattern.
Additionally, DNA sequencing was performed for a subset of 642 samples from 114 IMRT patients. In this cohort we observed that genomic heterogeneity was also indicative of a higher risk of recurrence. Combined, these two measurements of heterogeneity also identified a group of patients with half the median time to recurrence compared to the rest of the cohort (5.6 vs 11.5 years).
This study is a strong indicator that the combination of genomics and AI-aided histopathology in clinical trials have the potential to the identify of novel clinical biomarkers and improve treatment for higher risk patients.
Nikolas Stathonikos
UMC Utrecht
4 July 2022
Title: TBC
Abstract: TBC