Abstract:The placenta forms a critical barrier to infection through pregnancy, labor and, delivery. Inflammatory processes in the placenta have short-term, and long-term consequences for offspring health. Digital pathology and machine learning can play an important role in understanding placental inflammation, and there have been very few investigations into methods for predicting and understanding Maternal Inflammatory Response (MIR). This work intends to investigate the potential of using machine learning to understand MIR based on whole slide images (WSI), and establish early benchmarks. To that end, we use Multiple Instance Learning framework with 3 feature extractors: ImageNet-based EfficientNet-v2s, and 2 histopathology foundation models, UNI and Phikon to investigate predictability of MIR stage from histopathology WSIs. We also interpret predictions from these models using the learned attention maps from these models. We also use the MIL framework for predicting white blood cells count (WBC) and maximum fever temperature ($T_{max}$). Attention-based MIL models are able to classify MIR with a balanced accuracy of up to 88.5% with a Cohen's Kappa ($\kappa$) of up to 0.772. Furthermore, we found that the pathology foundation models (UNI and Phikon) are both able to achieve higher performance with balanced accuracy and $\kappa$, compared to ImageNet-based feature extractor (EfficientNet-v2s). For WBC and $T_{max}$ prediction, we found mild correlation between actual values and those predicted from histopathology WSIs. We used MIL framework for predicting MIR stage from WSIs, and compared effectiveness of foundation models as feature extractors, with that of an ImageNet-based model. We further investigated model failure cases and found them to be either edge cases prone to interobserver variability, examples of pathologist's overreach, or mislabeled due to processing errors.

What problem does this paper attempt to address?

The key problem that this paper attempts to solve is how to use machine - learning techniques, especially the Multiple Instance Learning (MIL) framework based on deep learning, to identify the Maternal Inflammatory Response (MIR) and histological chorioamnionitis from Whole Slide Images (WSI) of the placental membrane. Specifically, the main objectives of the study include: 1. **Predicting MIR stages**: By using the MIL model to classify different stages of MIR (MIR0,1 and MIR2,3) from WSI, and evaluating the performance of different feature extractors (such as the ImageNet pre - trained model EfficientNet - v2s, the pathology - based models UNI and Phikon). 2. **Exploring the prediction of white blood cell count and maximum body temperature**: As an exploratory analysis, the study also attempts to predict white blood cell count (WBC) and maximum body temperature ($T_{max}$) from WSI to evaluate the association between these physiological indicators and MIR. ### Research Background The placenta acts as a barrier during pregnancy and childbirth to prevent infection. The inflammatory process in the placenta has short - term and long - term effects on the health of offspring. Digital pathology and machine learning can play an important role in this regard, but currently there are few studies on methods for predicting and understanding MIR. Therefore, this study aims to fill this gap and establish an early benchmark. ### Method Overview - **Dataset**: The study used 3,385 placental samples collected in a single institution between 2010 and 2024. - **Feature Extraction**: Three different feature extractors were used: EfficientNet - v2s (based on ImageNet), UNI, and Phikon (pathology - based models). - **Model Architecture**: An MIL model based on the attention mechanism was adopted, which can assign weights to each image block according to its importance. - **Evaluation Metrics**: For the MIR classification task, metrics such as AUROC, balanced accuracy, Matthews correlation coefficient (MCC), and Cohen’s Kappa ($\kappa$) were used; for the WBC and $T_{max}$ prediction tasks, metrics such as mean squared error (MSE), mean absolute error (MAE), and $R^{2}$ score were used. ### Main Findings - **MIR Classification Performance**: The MIL models based on pathology - based models (UNI and Phikon) outperformed the ImageNet - based EfficientNet - v2s in the MIR classification task. Among them, Phikon achieved the highest balanced accuracy (88.5%) and Cohen’s Kappa (0.772). - **WBC and $T_{max}$ Prediction**: Although there is a certain correlation, the prediction effect is relatively weak, especially for WBC prediction. ### Conclusion The study shows that the MIL framework based on pathology - based models has high accuracy in predicting MIR stages and is superior to models pre - trained on general - purpose image datasets. In addition, although the correlation of WBC and $T_{max}$ prediction is weak, it still shows statistical significance, providing a direction for further research. Through this study, the authors hope to provide powerful tools and technical support for a more in - depth understanding of MIR and its clinical significance in the future.

Machine learning identification of maternal inflammatory response and histologic choroamnionitis from placental membrane whole slide images

Quantitative Modeling to Characterize Maternal Inflammatory Response of Histologic Chorioamnionitis in Placental Membranes

Deep Learning for Fetal Inflammatory Response Diagnosis in the Umbilical Cord

Identification of Preeclamptic Placenta in Whole Slide Images Using Artificial Intelligence Placenta Analysis

Radiomics Analysis of Placenta on T2WI Facilitates Prediction of Postpartum Haemorrhage: A Multicentre Study.

In Vivo Assessment of the Placental Anatomy and Perfusion in a Mouse Model of Intrauterine Inflammation.

Towards Deep Cellular Phenotyping in Placental Histology

580: Acute MRI Changes in Mouse Placentas Exposed to Intrauterine Inflammation

Deep learning-based image analysis methods for brightfield-acquired multiplex immunohistochemistry images

Machine Learning for the Prediction of Surgical Morbidity in Placenta Accreta Spectrum

Precision Medicine for Chronic Endometritis: Computer-Aided Diagnosis Using Deep Learning Model

Identification and Verification of Diagnostic Biomarkers Based on Mitochondria-Related Genes Related to Immune Microenvironment for Preeclampsia Using Machine Learning Algorithms

Prenatal Prediction and Typing of Placental Invasion Using MRI Deep and Radiomic Features

Non-invasive determination of gene expression in placental tissue using maternal plasma cell-free DNA fragmentation characters

Placental Barrier-on-a-Chip: Modeling Placental Inflammatory Responses to Bacterial Infection

Development and validation of preeclampsia predictive models using key genes from bioinformatics and machine learning approaches

GestAltNet: aggregation and attention to improve deep learning of gestational age from placental whole-slide images

Predictive potential of various plasma inflammation-, angiogenesis-, and extracellular matrix remodeling-associated mediators for intra-amniotic inflammation and/or microbial invasion of the amniotic cavity in preterm labor

Identification and characterization of plasma proteins associated with intra-amniotic inflammation and/or infection in women with preterm labor

Prediction and verification of benignancy and malignancy of pulmonary nodules based on inflammatory related biological markers

A deep learning model for lymph node metastasis prediction based on digital histopathological images of primary endometrial cancer