Anne-Marie Dingemans

Erasmus MC

In short, Anne-Marie Dingemans' research entails the following: 

Histogenomic biomarker identification to improve neuroendocrine lung tumor diagnostics on biopsies, using multiplex immunohistochemistry and artificial intelligence assisted histomorphological classification

'Pulmonary neuroendocrine tumours (pNETs) comprise four rare lung tumour types, which share some histological and phenotypical (protein immunostaining) features. However, they are characterized by divergent biologic behaviour. Improvement in sub-classification of pNETS on biopsy specimens will facilitate personalised treatment and improve patient outcome. In this study we aim to substantially improve pNET diagnosis and prognosis by application of artificial intelligence (AI)-driven morphology detection informed by applying genomic single slide (multiplex) IHC markers.

Aim 1: To predict the recurrence of disease after surgery in carcinoid pNETs on pre-operative biopsy specimens, we will use a combined AI-histogenomic approach using H&E stained and multiplex IHC stained slides.

Aim 2: To improve the identification of LCNEC on biopsy specimens, an AI model will be developed, which can distinguish neuroendocrine class from non-neuroendocrine class in digital images. This model will be trained for neuroendocrine class detection. The combined morphology and IHC AI model will then be validated on a separate pre-operative biopsy specimen cohort with matched surgical resections as ground truth.

Aim 3: To explore whether LCNEC is a homogeneous morphological entity or can be morphologically subdivided into carcinoid/large cell/SCLC subtypes, we will use cluster analysis of all cases above, plus additional SCLC. These subtypes will be further correlated to known genomic LCNEC subtypes.

This project is a collaboration between Erasmus MC Rotterdam (prof Anne-Marie Dingemans, dr Jan von der Thusen, dr Andrew Stubbs, dr Yunlei Li) and Maastricht UMC+ Maastricht (prof Ernst-Jan Speel, dr Jules Derks).'

Brief summary of progress / results January 2024

In this progress report we show that we have reached significant progress in all three work packages.
In WP1 we have nearly completed and digitalised all required tumor cohorts (carcinoid, SCLC and NSCLC). The manual tumor annotation is work in progress and for the LCNEC cohort we expect to receive the last samples in Q1 2024.

The focus of WP2 is immunohistochemical multiplex analysis and tumor annotation. The multiplex analysis of the carcinoid TMA cohort is completed, although the optimalisation of the multiplex IHC panel was challenging. Singleplex IHC analyses of the TMA revealed promissing results and will be used as ground truth for the AI experiments.

In WP3 we have gained some preliminary results in AI driven histogenomic predictions modelling. The annotation of the TMA H&E specimen is completed, and prediction models for carcinoid recurrence are promising with a balanced accuracy of 0.80 by validation. For prediction of morphological subtypes of LCNEC a balanced accuracy of 0.79 and 0.64 was reached in patient analysis

Brief summary of progress / results February 2023

The requested PhD students and post-doc are appointed. The most important goal of the first year was to expand our tumor cohorts. We collected additional tumor samples and constructed TMA's of the LCNEC/Carcinoid cohorts. We set up a collaboration with a group in Vienna in order to expand our SCLC cohort. The multiplex immunofluorescent panel is established and evaluated on the carcinoid TMA and currently optimalisation is ongoing, next to comparing the results with single plex chromogenic staining. We developed an AI framework for recurrence prediction in lung carcinoid TMA. More specifically, H&E TMA from 139 patients (278 cores) were pre-processed and a pre-trained VGG16 Net model was trained and tested in a 5-fold cross-validation setting. We hope to present the first analysis of the carcinoid cohort on European Society of Pathology conference in September 2023.