ILC is the second most common specific type of breast cancer
ILC presents specific challenges across the clinical pathway, from detection and pathological classification to treatment response and disease monitoring. These challenges arise from its distinctive growth patterns, imaging characteristics and molecular biology, and can complicate timely and accurate clinical decision-making
Inclusion rate of ILC patients in
clinical trials is 2–26%
Current screening modalities
are debated for ILC patients
Integrated use of morphology, IHC,
and molecular findings is debated
Pathologists use a combination of tissue stains and biomarkers to identify and characterise invasive lobular carcinoma. These stains make different features of the tumour visible, from its growth pattern and cell morphology to the expression of proteins that help confirm the diagnosis and guide treatment. The images below show some of the key markers used in ILC assessment.
The standard stain used to assess tumour growth pattern, cytological morphology and tumour cell proliferation. It is essential to identify ILC growth patterns and distinguish classic from non-classic patterns (cytological and architectural). H&E is crucial to define tumour grade.
Dysplastic tumour cells loosely arranged or in single files are sometimes visible with optically empty intracytoplasmic inclusions. A normal duct is also visible.
The duct shown on the left upper corner show ductal carcinoma in situ (DCIS) with membranous intense brown staining (honeycomb pattern), while the linearly arranged invasive tumour cells show lack of E-cadherin expression. Nuclei are stained with haematoxylin (blue).
Likewise, E-cadherin, beta-catenin shows a honeycomb brown staining in the duct with DCIS on the left. The tumour cells show loss of expression, or sometimes weak cytoplasmic expression. Fibroblasts in the background may also show weak, linear expression. Nuclei are stained blue with haematoxylin.
Strong and diffuse nuclear ER expression in brown is observed in both DCIS and ILC. ER+ is observed in the vast majority of ILC, indicating eligibility for endocrine therapy.
Strong and diffuse nuclear PR expression in brown is observed in DCIS, while in ILC, PR expression is weaker and less diffuse. Loss of PR expression may indicate a worse prognosis or reduced ER signalling dependence in hormone receptor-positive breast cancers.
Both DCIS and ILC show faint fragmented membranous HER2 expression, being classified as a 1+ staining pattern and therefore being ineligible for anti-HER2 targeted therapy.
Despite advances in breast cancer diagnostics and artificial intelligence, information relevant to ILC remains distributed across different data types, clinical centres and stages of care. M4GIC-ILC addresses this fragmentation by integrating multimodal and multi-site data within a unified, ILC-specific AI framework.
To connect high-quality data with clinically oriented GenAI, M4GIC-ILC aims to start with data collection and harmonisation, before moving on to representation learning, predictive modelling and, ultimately, a multi-agent system designed to support clinicians through explainable and interactive decision support.
M4GIC-ILC aims to integrate clinical, pathology, imaging, and genomic data from multiple European centres. Upon data harmonisation, different AI models trained for specific tasks, such as the generation of in silico IHC images, are guided by three main master agents. The master agents will work together to ultimately help clinicians by providing diagnosis and staging support, relapse risk predictions, and personalised treatment selections.
Over three years, key milestones will guide the development of the data infrastructure, AI models and multi-agent system towards the final goal of an integrated clinical tool.
