What it is
TITAN (Transformer-based pathology Image and Text Alignment Network) is a whole-slide foundation model pretrained on 335,645 whole-slide images via visual self-supervised learning plus vision-language alignment with pathology reports and 423,122 synthetic captions. Without any fine-tuning or clinical labels, it extracts general-purpose slide representations and generates pathology reports. Across linear probing, few-shot and zero-shot classification, rare cancer retrieval, cross-modal retrieval, and report generation, it outperforms prior region-of-interest and slide foundation models.
Why it matters
Computational pathology has been limited at the patient and slide level because disease-specific cohorts, especially for rare conditions, rarely have enough labeled data to train on. TITAN targets exactly that gap: it transfers to rare cancer retrieval and cancer prognosis in a zero-shot or few-shot regime, so the hardest, data-scarce clinical scenarios do not each require their own labeled training set.
Underlined numbers link to their source. Every metric and quoted figure is listed under Sources and data below.
Filed underfoundation model, computational pathology, whole-slide imaging, vision-language, cancer diagnosis