Abstract
Spatial transcriptomics preserves tissue context while profiling gene expression, but its high cost and experimental complexity limit large-scale use. This has motivated methods that predict spatial gene expression directly from H\&E histopathology images. However, most existing approaches are trained on individual datasets with fixed gene sets, limiting the joint use of heterogeneous projects.
We propose MINT-ST, a multi-dataset pretraining framework for histology-based spatial gene expression prediction. MINT-ST constructs a global output space from the union of project-specific highly variable genes, jointly pretrains a shared model across multiple projects, and fine-tunes it for each target gene panel and expression distribution. A Transformer integrates the target spot, neighboring spots, relative positions, and species and organ metadata. Training combines variance-normalized regression with a local Pearson correlation loss to learn both expression values and spatial patterns.
Across ten human and mouse SpaRED datasets, MINT-ST achieved the best average performance across all evaluation metrics. Ablation studies verified the contributions of spatial context, metadata, correlation-based supervision, multi-project pretraining, and project-specific fine-tuning.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MISO_009.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=z5DZvwOFBJ
BibTex
@InProceedings{ByeKeu_MINTST_MICCAISAT2026,
author = { Byeon, Keunho AND Kwak, Jin Tae},
title = { { MINT-ST: Multi-Dataset Integration and Pretraining for Histology-Based Spatial Transcriptomics Gene Expression Prediction } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17252},
month = {pending},
page = {pending}
}
