Development of a Deep Learning model Tailored for HER2 Detection in Breast Cancer to aid pathologists in interpreting HER2-Low cases

Bannier, P.-A.; Broeckx, G.; Herpin, L.; Dubois, R.; Van Praet, L.; Maussion, C.; Deman, F.; Amonoo, E.; Mera, A.; Timbres, J.; Gillett, C.; Sawyer, E.; Gazinska, P.; Ziolkowski, P.; Lacroix-Triki, M.; Salgado, R.; Irshad, S.

doi:10.1101/2024.07.01.601397

bioRxiv · 10.1101/2024.07.01.601397

Development of a Deep Learning model Tailored for HER2 Detection in Breast Cancer to aid pathologists in interpreting HER2-Low cases

Bannier, P.-A.·Broeckx, G.·Herpin, L.·Dubois, R.·Van Praet, L.·Maussion, C.·Deman, F.·Amonoo, E.·Mera, A.·Timbres, J.·Gillett, C.·Sawyer, E.·Gazinska, P.·Ziolkowski, P.·Lacroix-Triki, M.·Salgado, R.·Irshad, S.

View original source ↗Export metadata ↓

Abstract

IntroductionOver 50% of breast cancer cases are "Human epidermal growth factor receptor 2 (HER2) low breast cancer (BC)", characterized by HER2 immunohistochemistry (IHC) scores of 1+ or 2+ alongside no amplification on fluorescence in situ hybridization (FISH) testing. The development of new anti-HER2 antibody-drug conjugates (ADCs) for treating HER2-low breast cancers illustrates the importance of accurately assessing HER2 status, particularly HER2-low breast cancer. In this study, we evaluated the performance of a deep learning (DL) model for the assessment of HER2, including an assessment of the causes of discordances of HER2-Null between a pathologist and the DL model. We specifically focussed on aligning the DL model rules with the ASCO/CAP guidelines, including stained cells staining intensity and completeness of membrane staining. MethodsWe trained a DL model on a multi-centric cohort of breast cancer cases with HER2- immunohistochemistry scores (n=299). The model was validated on 2 independent multi- centric validation cohorts (n=369 and n=92), with all cases reviewed by 3 senior breast pathologists. All cases underwent a thorough review by three senior breast pathologists, with the ground truth determined by a majority consensus on the final HER2 score among the pathologists. In total, 760 breast cancer cases were utilized throughout the training and validation phases of the study. ResultsThe models concordance with the ground truth (ICC = 0.77 [0.68 - 0.83]; Fisher P = 1.32e-10) is higher than the average agreement among the 3 senior pathologists (ICC = 0.45 [0.17 - 0.65]; Fisher P = 2e-3). In the two validation cohorts, the DL model identifies 95% [93%- 98%] and 97% [91% - 100%] of HER2-low and HER2-positive tumors respectively. Discordant results were characterized by morphological features such as extended fibrosis, a high number of tumor-infiltrating lymphocytes, and necrosis, whilst some artifacts such as non- specific background cytoplasmic stain in the cytoplasm of tumor cells also cause discrepancy. ConclusionDeep learning can support pathologists interpretation of difficult HER2-low cases. Morphological variables and some specific artifacts can cause discrepant HER2-scores between the pathologist and the DL Model.

Explore related subjects

pathology Preprint

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Bannier, P.-A., Broeckx, G., Herpin, L., Dubois, R., Van Praet, L., Maussion, C., Deman, F., Amonoo, E., Mera, A., Timbres, J., Gillett, C., Sawyer, E., Gazinska, P., Ziolkowski, P., Lacroix-Triki, M., Salgado, R., Irshad, S.. 2024-07-03. Development of a Deep Learning model Tailored for HER2 Detection in Breast Cancer to aid pathologists in interpreting HER2-Low cases. https://doi.org/10.1101/2024.07.01.601397

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

Dysregulated Platelet GPIb alpha - VWF Signalling in Abdominal Aortic Aneurysm formation and Progression

Background: Platelets are critical drivers of thrombo-inflammatory responses in different cardiovascular diseases. Abdominal aortic aneurysm (AAA) is a progressive, life-threatening vascular disorder mainly characterised by chronic inflammation, extracellular matrix degradation, and the formation of a platelet-rich intraluminal thrombus (ILT). Experimental and clinical evidence identified platelets as main players in AAA pathology as evidenced by elevated platelet activation and procoagulant activity that critically contribute to AAA progression. Methods: The present study investigated the contribution of glycoprotein (GP)Ib alpha, the von Willebrand factor (VWF)-binding subunit of the platelet GPIb-IX-V complex, to AAA initiation and progression in experimental AAA using the ePPE mouse model and in patients. Results: Genetic ablation of platelet GPIb alpha significantly attenuated early aneurysm expansion in experimental AAA, indicating a critical role for GPIb alpha during the initial stages of aneurysm development. This initial effect was compensated at later time points showing no differences in aneurysm progression between groups. Notably, genetic deletion of GPIb alpha induced a constitutively hyperactive platelet phenotype already in naive mice that was further amplified during experimental AAA. This elevated platelet hyperactivity was mainly due to increased GPVI activation of platelets 28 days post-surgery. To assess the clinical relevance, spatial profiles of human ILT specimens from patients with AAA were analysed. In the ILT, we detected a highly compartmentalised distribution of GPIb alpha and VWF with pronounced enrichment within the luminal layer. In parallel, circulating VWF activity as well as platelet surface expression of GPIb alpha were significantly increased in patients with AAA. Conclusion: Collectively, these findings identify a dysregulated GPIb alpha-VWF axis in human AAA pathology, mainly characterised by enhanced platelet GPIb alpha surface expression and increased activity of circulating VWF.

pathology↗

bioRxiv2026

OmiCoreTumorDetector: an open, molecularly validated model for mapping tumour regions in colorectal cancer H&E sections

Defining tumour regions on haematoxylin and eosin (H&E) sections is a routine first step in spatial-omics studies, yet it is usually done by hand and is difficult to reproduce. We present OmiCoreTumorDetector, an openly licensed model that maps tumour-enriched regions in colorectal cancer (CRC) H&E sections and exports them as QuPath-compatible annotations. The released model (omicore-tumordetector-crc-he-v0.1) is an ensemble of three convolutional classifiers trained on 100,000 public tissue tiles, combined with Macenko stain normalisation at inference. During development we found that the main obstacle to reuse was calibration under stain-domain shift rather than discrimination: a single model kept an area under the ROC curve (AUROC) of 0.955 on unseen slides while its sensitivity at the conventional 0.5 threshold fell to 0.48. Training on non-normalised tiles raised tumour AUROC on an independently collected tile set from 0.836 to 0.992, and normalising at inference reduced false-positive tumour area in normal-adjacent tissue by 16- to 26-fold. On five 10x Visium HD CRC sections that share no material with the training data, the released model called 27.7-48.5% of tissue as tumour in three carcinoma sections and 0.08% and 1.82% in two normal-adjacent sections, exporting no tumour region from either normal section. On the carcinoma section with matched single-cell-resolution transcriptomics, agreement with transcriptome-derived tumour-cell identities reached an AUROC of 0.985 (95% spatial-block bootstrap CI 0.975-0.993). The image model never observes gene expression, so this is orthogonal evidence. The model localises tumour-enriched regions at 112 um resolution; it does not identify individual malignant cells and has not yet been validated across scanners, institutions or histological variants. Code, weights and evaluation are released under Apache-2.0 and installable with pip install omicoretumordetector.

pathology↗

bioRxiv2026

Thyroid Dysfunction in Male Patients at Asia Med Laboratory, Herat, Afghanistan July 2021-Jan 2022

Objective: Hyperthyroidism and hypothyroidism related to iodine deficiency are major public health concerns in Afghanistan. This study aimed to assess the frequency of thyroid dysfunction among male patients referred for thyroid testing and its association with age, and to examine monthly trends in thyroid dysfunction at Asia Med Laboratory in Herat, Afghanistan, from July 2021 to January 2022. Methods: A retrospective analysis was conducted on 250 male patients aged 0-69 years. We measured Serum TSH, total T4, and total T3 levels, and thyroid status was classified using age specific reference ranges. In addition, the frequency of thyroid dysfunction was analyzed across age groups with monthly trends of thyroid state. Results: Overall, the euthyroid state consisted of 69.2% of participants, 24.8% with overt hypothyroidism, 3.2% with overt hyperthyroidism, and 2.8% with subclinical hyperthyroidism. Thyroid status differed significantly by age (p = 0.0135), with hypothyroidism increasing in older age groups and reaching its highest proportion among men aged 60-69 years (55.6%). Euthyroidism predominated in patients aged 10-39 years, while hyperthyroidism across age groups remained relatively infrequent. After September 2021, a threefold increase was observed in the total number of male patients referred for thyroid testing. During this period, the proportion of hyperthyroidism increased slightly, whereas hypothyroidism cases declined. Conclusion: In conclusion, hypothyroidism was more frequent with older age. The rise in absolute case numbers after September 2021 likely reflects increased patient referrals, underscoring the need for ongoing monitoring of thyroid function. The study may assist in the early management of thyroid disorders and in reducing their complications.

pathology↗