Publications

Whole Slide Images (WSIs) are a core data modality in computational pathology, yet their gigapixel resolution requires weakly supervised approaches such as Multiple Instance Learning (MIL) for prognostic modeling. While recent advances in representation learning have introduced domain-specific and foundation models for histopathology, it remains unclear how the choice of patch-level embedding influences survival prediction performance when combined with different MIL architectures and applied across heterogeneous cancer cohorts. Systematic evaluations addressing this gap are still limited. In this work, we present a comprehensive benchmark aimed at deriving practical guidelines for MIL-based Overall Survival (OS) prediction from WSIs. We compare four representative tile embedding strategies (ResNet50, ProvGigaPath, UNI, and CONCH) across five The Cancer Genome Atlas (TCGA) cohorts: bladder urothelial carcinoma (BLCA), breast invasive carcinoma (BRCA), colon adenocarcinoma (COAD), head and neck squamous cell carcinoma (HNSC), stomach adenocarcinoma (STAD) and one dataset from the Clinical Proteomic Tumor Analysis Consortium (CPTAC), clear cell renal cell carcinoma (ccRCC). Patch-level features are aggregated using three state-of-the-art MIL survival models: Attention-Based MIL (ABMIL), Transformer-based MIL (TransMIL), and Dual-Stream MIL (DSMIL). Furthermore, the analysis is extended to include a comparison with two state-of-the-art slide-level encoders, TITAN and the ProvGigaPath slide encoder, utilizing a Cox Proportional Hazard (CPH) model as the prediction head to benchmark patch-aggregation approaches against end-to-end slide representations. Prognostic performance is assessed using the concordance index (c-index), with interpretability evaluated through risk stratification analysis. Our results show that embedding choice has a substantial and consistent impact on OS prediction accuracy, robustness, and interpretability across cancer types, with domain-specific and foundation models outperforming conventional convolutional baselines. These findings provide evidence-based practical guidelines for designing robust and generalizable WSI-based survival prediction pipelines in computational pathology.

Drawing analogies between two pairs of entities in the form of A:B::C:D (i.e. A is to B as C is to D) is a hallmark of human intelligence, as evidenced by sufficient findings in cognitive science for the last decades. In recent years, this property has been found far beyond cognitive science. Notable examples are word2vec and GloVe models in natural language processing. Recent research in computer vision also found the property of analogies in the feature space of a pretrained ConvNet feature extractor. However, analogy mining in the semantic space of recent strong foundation models such as CLIP is still understudied, despite the fact that they have been successfully applied to a wide range of downstream tasks. In this work, we show that CLIP possesses the similar ability of analogical reasoning in the latent space, and propose a novel strategy to extract analogies between pairs of images in the CLIP space. We compute all the difference vectors of a pair of any two images that belong to the same class in the CLIP space, and employ k-means clustering to group the difference vectors into clusters irrespective of their classes. This procedure results in cluster centroids representative of class-agnostic semantic analogies between images. Through extensive analysis, we show that the property of drawing analogies between images also exists in the CLIP space, which are interpretable by humans through a visualisation of the learned clusters.