Expansion microscopy (ExPath) enables nanoscale resolution of tissue architecture using conventional microscopes, offering a powerful alternative to traditional histopathology. In this thesis, we present a deep learning pipeline that leverages ExPath imaging combined with a biologically informed, four-channel multiplexed staining panel: DAPI, TelC, CENPB, and WGA to classify tissue types and predict chemotherapy response in muscle-invasive bladder cancer (MIBC). We propose that nuclear morphology, when captured at high resolution and enriched by chromatin and membrane-specific markers, contains sufficient information to compete with H&E and generalize across diagnostic and prognostic tasks. To test this hypothesis, we construct a preprocessing pipeline that transforms 16-bit 4-channel TIFF WSIs into normalized, pseudo-RGB 1024x1024 patches compatible with ImageNet-pretrained models. We evaluate multiple architectures (ResNet34, ResNet50, ViT-tiny, EfficientNet) and demonstrate that ResNet-based models trained on ExPath outperform simulated non-ExPath baselines and DAPI-only variants by a significant margin. Through controlled ablation experiments, we quantify the contribution of each channel and find that multiplexing substantially boosts classification accuracy. Our models achieve 89.52% tissue classification accuracy and 0.9 ROC-AUC for drug response prediction. Furthermore, we observe cross-cancer generalizability when applying MIBC-trained models to lung carcinoma ExPath images. This work establishes the feasibility of compact, multiplexed, ExPath-driven classification pipelines as a viable alternative to costly multi-modal diagnostics. It offers an early step toward a DAPI-first foundation model for computational pathology, with potential to scale across cancer types and tissue conditions using minimal staining and high-content imaging.

47 pages

Thesis Committee:
Russell S. Schwartz (Chair)
Min Xu

Srinivasan Seshan, Head, Computer Science Department
Martial Hebert, Dean, School of Computer Science

Return to: SCS Technical Report Collection
School of Computer Science

This page maintained by reports@cs.cmu.edu