With a high risk of relapse and death, and a poor or absent response to therapeutics, the triple-negative breast cancer (TNBC) subtype is particularly challenging, especially in patients who cannot achieve a pathological complete response (pCR) after neoadjuvant chemotherapy (NAC). Although the tumor microenvironment (TME) is known to influence disease progression and the effectiveness of therapeutics, its predictive and prognostic potential remains uncertain. This work aimed to define the residual TME profile after NAC of a retrospective cohort with 96 TNBC patients by immunohistochemical staining (cell markers) and chromogenic in situ hybridization (genetic markers). Kaplan–Meier curves were used to estimate the influence of the selected TME markers on five-year overall survival (OS) and relapse-free survival (RFS) probabilities. The risks of each variable being associated with relapse and death were determined through univariate and multivariate Cox analyses. We describe a unique tumor-infiltrating immune profile with high levels of lymphocytes (CD4, FOXP3) and dendritic cells (CD21, CD1a and CD83) that are valuable prognostic factors in post-NAC TNBC patients. Our study also demonstrates the value of considering not only cellular but also genetic TME markers such as MUC-1 and CXCL13 in routine clinical diagnosis to refine prognosis modelling.

Triple-negative breast cancer (TNBC) is particularly challenging due to the weak or absent response to therapeutics and its poor prognosis. The effectiveness of neoadjuvant chemotherapy (NAC) response is strongly influenced by changes in elements of the tumor microenvironment (TME). This work aimed to characterize the residual TME composition in 96 TNBC patients using immunohistochemistry and in situ hybridization techniques and evaluate its prognostic implications for partial responders vs. non-responders. Compared with non-responders, partial responders containing higher levels of CD83+ mature dendritic cells, FOXP3+ regulatory T cells, and IL-15 expression but lower CD138+ cell concentration exhibited better OS and RFS. However, along with tumor diameter and positive nodal status at diagnosis, matrix metalloproteinase-9 (MMP-9) expression in the residual TME was identified as an independent factor associated with the impaired response to NAC. This study yields new insights into the key components of the residual tumor bed, such as MMP-9, which is strictly associated with the lack of a pathological response to NAC. This knowledge might help early identification of TNBC patients less likely to respond to NAC and allow the establishment of new therapeutic targets.

Tumor-infiltrating lymphocytes (TILs) are prognostic biomarkers in breast cancer (BC), particularly in HER2-positive and triple-negative subtypes. Assessment follows the international guidelines, in which pathologists evaluate whole hematoxylin and eosin (H&E)-stained slides while integrating representative regions of the invasive tumor. However, manual region selection can be labor-intensive, subjective, and may introduce variability, particularly across consecutive tissue sections. Automated region of interest (ROI) registration may mitigate this limitation, yet its impact on longitudinal TIL scoring has not been systematically evaluated. Here, we introduce three ROI registration strategies (direct, intermediate, and serial with/without quality control) and present an automated framework validated for consistent TIL scoring and clinical relevance in predicting relapse. We analyzed 104 invasive BC cases, each with 12 consecutive H&E slides. A pathologist annotated ROIs on both the first and twelfth slides. We registered these ROIs using the proposed strategies. We then evaluated them with performance metrics, including Intersection over Union (IoU), Dice Similarity Coefficient (DSC), failure rate, and execution time. Two pathologists scored TILs on manual and automated ROIs. We assessed longitudinal consistency between the first manual ROI and the twelfth slide's corresponding ROIs. We also tested whether the association between TIL score and patient relapse outcomes was preserved. The direct registration strategy achieved the highest geometric accuracy (mean IoU = 0.650, DSC = 0.769), with < 1% failure rate and ∼5.8 s execution time. TIL scores for both manual and automated ROIs closely matched (P = 0.84), showing strong agreement (Spearman's ρ = 0.923, ICC = 0.936, CCC = 0.915, Cohen's κ = 0.786). Longitudinal analyses showed no significant variability across distant sections (P = 0.79 and P = 0.62). TIL concentrations differed significantly between patients with and without relapse, and this association was preserved on both the first and twelfth slides (P < 0.05). We present the first automated framework for longitudinal ROI registration to support TIL scoring in BC. By reducing manual effort and variability, the framework supports scalable evaluation of immune biomarkers across tissue depth while preserving clinically relevant signals, supporting precision immuno-oncology applications.

Histopathology slide digitization introduces scanner-induced domain shift that can significantly impact computational pathology models based on deep learning methods. In the state-of-the-art, this shift is often characterized at a broad scale (slide-level or dataset-level), but not at the patch level, which limits our comprehension of the impact of localized tissue characteristics on model accuracy. To address this challenge, we present a domain shift analysis framework based on UWarp, a novel registration tool designed to accurately align histological slides scanned under varying conditions. UWarp employs a hierarchical registration approach, combining global affine transformations with fine-grained local corrections to achieve robust tissue patch alignment. We evaluate UWarp using two private datasets, CypathLung and one from the BosomShield project, containing whole slide images scanned by multiple devices. Our experiments show that UWarp outperforms existing open-source registration methods, achieving a median target registration error (TRE) below 4 pixels (<1 μm at 40× magnification), compared to 21 pixels (~5 μm at 40×) for the best state-of-the-art algorithm, while significantly reducing computational time. Additionally, we perform a localized characterization of scanner-induced domain shift by analyzing the predictions of a deep learning model for breast cancer pathological response. Accurate patch-level alignment between scanners is ensured using UWarp. Our analysis reveals that prediction variability correlates with tissue versus background ratio and nuclei density at patch level, highlighting the importance of accounting for localized domain shift to improve model robustness and adaptation strategies in computational pathology.

Metastasis occurs in nearly 1 out of 3 breast cancer (BC) patients and significantly reduces survival rates, particularly in cases of distant metastases. As most distant metastases develop after diagnosis (i.e., recurrence) and remain incurable, there is a critical need for prognostic biomarkers to assess recurrence risk. Multimodal data analysis has emerged as a promising approach to integrate diverse information, offering a more comprehensive perspective. This study introduces the Histology HSI-BC (hyperspectral imaging - breast cancer) Recurrence Database, the first publicly accessible multimodal database designed to advance BC distant recurrence prediction. The database comprises 47 histopathological whole-slide images, 677 hyperspectral (HS) images, and clinical and demographic data from 47 BC patients, of whom 22 (47%) experienced distant recurrence over a 12-year follow-up. Histopathological slides were digitized using a whole-slide scanner and annotated by expert pathologists, while HS images were acquired with an HS camera coupled to a bright-field microscope. This database provides a promising resource for studying BC recurrence prediction and personalized treatment strategies by integrating the aforementioned multimodal data.