AI-based Prognostic Factors Identification in Early-stage Invasive Lobular Breast Carcinoma

Overview

Invasive lobular breast carcinoma (ILC), the second most common breast cancer type, represents approximately 15% of all breast cancer cases. Improvement in diagnostic modalities as well as growing evidence that hormone replacement therapy is associated with higher risk of both invasive lobular carcinoma (ILC) and invasive ductal-lobular mixed carcinoma (IDLC) have increased ILC incidence during last years. Most risk factors for ILC, which is more frequently diagnosed in women over the age of 50, are related to increased hormone exposure, including earlier menarche, late menopause, late age at first birth, oral contraceptives, menopausal hormone therapy. Lifestyle factors, like alcohol intake, Western diet and postmenopausal obesity are also associated with lobular breast cancer. Along with environment and lifestyle factors, genetic predisposition with germline mutations commonly found in CDH1 and FOXA1 are significant in ILC development. Other implicated genes include PIK3CA, PTEN, AKT1, and GATA3, while more than 50% of ILC harbor mutations in ERBB2, ESR1, FGF, or NF1, conferring endocrine therapy resistance. Interestingly, BRCA1 mutations are less frequent in ILC than in IDC, with similar frequencies of BRCA2, TP53, and CHEK2. The predominant genetic mutation in CDH1 results in loss of expression of the cell-cell adhesion molecule E-cadherin, leading to a "discohesive" morphology of lobular cells. E-cadherin loss is commonly accompanied by abnormal expression of other parts of the cadherin-catenin complex, like membranous B-catenin and cytoplasmic p120-catenin. The characteristic "Indianfile" pattern of infiltration allows tumor cells to move through the extracellular matrix while causing a very little disruption of the underlying anatomic structures which frequently makes ILC difficult to detect through physical examination or imaging studies, like standard mammograms. Patients with lobular carcinomas are generally diagnosed at a more advanced stage compared with invasive ductal carcinomas, with larger tumor sizes and more frequent lymph node invasion. Because these tumors are more likely hormone receptor positive and HER2 negative, endocrine therapy is the widely preferred therapeutic approach. Combining endocrine treatments with CDK4/6 inhibitors have shown promising results in HR+ breast cancer, in early as well as in a metastatic setting. However, since endocrine resistance is common in estrogen receptor-positive breast cancer, new therapeutic approaches are needed. Recent studies have found bromodomain and extraterminal (BET) inhibition as well as FGFR inhibition, since ILC is known to have also FGFR-1 mutations causing resistance to BET inhibition, to be a potential therapeutic strategy for endocrine-resistant ILC cases. Other promising potential therapies rely on the role of mTOR inhibitors since mutations in the PI3K/Akt signaling pathway are the second most common in ILC and acquired resistance to hormonal therapies is linked to PI3K/AKT/mTOR activation. Targeted immune checkpoint inhibitors therapy is also being investigated in a subset of ILC with higher tumor-infiltrating lymphocytes (TILs) and PD-L1 expression. Finaly, since ILCs seem to have a significant number of cases which are now classified as HER2-low the use of antibody drug conjugates needs to be studied in these cases, preferably by designing multicentric randomized control trials. There is an urgent need to enhance our understanding of the clinicopathological and molecular features of invasive lobular breast cancer subtype with the goal of refining existing classifications and/or identifying potential biomarkers that could ultimately help improve therapeutic outcomes with established and novel therapies. Conclusively, the ILC less aggressive biological profile (strong hormone receptor positivity, low proliferative activity, lower histological grade) does not reflect a better long-term outcome. ILC overall biological and clinical features entail a cautious diagnostic and therapeutic approach and as mentioned before, the fact that ILC cells spread in single-file patterns rather than forming distinct masses make them difficult to detect. Digitalization and deep learning (DL) could definitely enhance ILC management. We intend to use a DL model in digitalized pathology slides and genomic data to improve ILC detection, predict recurrence risk, identify novel biomarkers and guide personalized treatments. Indeed, AI-driven models have been applied to whole-slide digital pathology images (WSIs) to predict CDH1 mutations from H\&E-stained slides. Moreover, a novel AI-derived tumour microenvironment risk score has been tested for long-term risk assessment supporting consideration of extended endocrine therapy in patients with ER+/HER2-, node-negative ILC.