PhD Scientific Days 2026

Pathological and Oncological Sciences 2.

Artificial Intelligence–Based Assessment of Stromal and Immune Microenvironment Features Predicts Response and Survival Following Neoadjuvant Chemotherapy in Breast Cancer

Előadó neve

Dr. Karancsi, Zsofia

Neptune code

opa9sv

Előadó munkahelye

Department of Pathology, Forensic and Insurance Medicine

Előadó telefonszáma

+36303147534

Előadó e-mail címe

karancsi.zsofia@stud.semmelweis.hu

Az előadás címe

Artificial Intelligence–Based Assessment of Stromal and Immune Microenvironment Features Predicts Response and Survival Following Neoadjuvant Chemotherapy in Breast Cancer

Szerző(k) neve és munkahelye

Zsofia Karancsi¹

1: Department of Pathology, Forensic and Insurance Medicine

Bemutatás módja

Szóbeli

Szekció

Pathological and Oncological Sciences 2.

Language of the presentation

English

Preferred session

Pathological and Oncological Sciences

Összefoglaló szövege

Introduction: Current biomarkers incompletely predict response to neoadjuvant chemotherapy (NAC) in breast cancer. Artificial intelligence (AI) and digital image analysis (DIA) enable rapid, objective assessment of tumor microenvironment features from H&E slides, supporting individualized risk stratification.

Patients and methods: We analyzed 202 digitized H&E core biopsies from breast cancer patients treated with NAC at Semmelweis University (2005–2025). Mean age was 53 years; pCR occurred in 83/202 patients (41%). Median follow-up was 58 months, with progression in 38 cases (19%). Subtypes included Luminal B1 (45), Luminal B2 (49), HER2 (31) and TNBC (77). Tumor–stroma ratio (TSR) and overall stroma ratio (OSR) were visually assessed. QuantCenter (3DHistech) differentiated tumor, stroma and background, while PathAI AISight generated 44 features, including stromal area, fibroblast and lymphocyte metrics.

Results: Visual and digital TSR/OSR assessments showed excellent concordance (ICC>0.9), with lower agreement for AI (ICC=0.75). Established biomarkers significantly predicted pCR, including Ki67 (OR=1.75, p<0.001), TILs (OR=2.06, p<0.001) and ER status (OR=0.34, p<0.001). TSR and OSR showed limited predictive value overall; however, higher OSR predicted lower pCR in TNBC (OR=0.54, p=0.017). PathAI immune-related metrics demonstrated the strongest associations, including stromal lymphocyte density (OR=2.27, p<0.001) and stromal lymphocyte-to-fibroblast ratio (OR=2.55, p<0.001). Fibroblast abundance predicted reduced response in TNBC (OR=0.53, p=0.021) but favorable response in Luminal B2 tumors (OR=2.23, p=0.03).
Routine core-biopsy biomarkers had no prognostic value for progression-free survival. In contrast, TSR and OSR were adverse prognostic factors across visual and digital assessments (HR=1.53–1.73, p=0.007–0.025), particularly in ER-negative and high-Ki67 tumors. Fibroblast proportion among stromal cells was also associated with worse PFS (HR=1.39, p=0.024). These associations remained significant in multivariable analyses.

Conclusions: AI- and DIA-derived stromal and immune biomarkers provide rapid, objective prediction of NAC response and prognosis in breast cancer. Immune-related features outperformed stromal metrics for pCR prediction, while stromal abundance retained prognostic significance for survival.

Ethical approval: BM/27896-3/2024

University

Semmelweis University

Supervisor

Prof. Anna Mária Tőkés, Prof. Janina Kulka

Publication of my abstract

I give consent to the publication of my abstract on the website of the congress.

phd.section.field

after finishing doctoral studies with absolutorium (PhD)

Kind

Szabad

Status

elfogadva

Accepted presentation method

szóbeli

Előadás fájl jóváhagyás

nem rendelkezett róla

Előadó

6893

Start

13:00

End

13:10