Truace
Subscribe

Knowledge-guided Multimodal Learning for Interpretable Prediction of Pathological Complete Response and Survival in Breast Cancer

Purpose Pretreatment estimation of pathological complete response (pCR) to neoadjuvant chemotherapy may support risk stratification and future response-adapted clinical studies in breast cancer. However, existing deep learning approaches often rely on "black-box" visual pattern recognition, suffering from poor interpretability, high data dependency, and limited generalizability across heterogeneous cohorts. Here, we present CLASP (Clinical Language-Aligned Slide Pathology), a knowledge-driven, multimodal framewo…

By the Truvace desk1 min read1 sourcePeer-reviewedTRV-2026-1658· cite & verify
Knowledge-guided Multimodal Learning for Interpretable Prediction of Pathological Complete Response and Survival in Breast Cancer
"Capillaritis-Acute hemorrhage - Milk allergy Case 107" by Pulmonary Pathology is licensed under CC BY-SA 2.0. To view a copy of this license, visit https://creativecommons.org/licenses/by-sa/2.0/.

In brief

Researchers built CLASP, a multimodal framework that merges whole slide image features, patient clinical data and 16 pathology text priors produced by an ensemble of large language models. It is trained to jointly predict pathological complete response to neoadjuvant chemotherapy and prognosis in breast cancer.

Across 1,486 patients from four cohorts, it scored higher AUC point estimates than 11 baseline methods. The authors describe it only as a candidate framework for pretreatment risk stratification, so its clinical usefulness is still to be shown, and the reported results come from retrospective validation.

Main points

  1. CLASP was validated on 1,486 patients from four independent cohorts across multiple medical centers.
  2. AUC was 0.881 in internal testing and 0.875 and 0.845 in two external validations.
  3. It had higher AUC point estimates than the 11 baseline methods evaluated.
  4. Few-shot performance remained competitive with as few as 8 training samples.
  5. Highlighted regions matched recognized features such as tumor-infiltrating lymphocytes and stromal architecture.

The gain

A model combining slide images, clinical data and language-model-generated pathology knowledge predicted complete response with AUC 0.845 to 0.881 across internal and external cohorts.

The rundown

The method targets common weaknesses of deep learning in pathology: black-box pattern recognition, poor interpretability, heavy data dependence and limited generalizability across heterogeneous cohorts. CLASP encodes slides with the CONCH foundation model and aligns visual biomarkers with semantic concepts.

A vision-language feature aggregation module and a multi-task consistency constraint let the model predict pCR and prognosis together. Interpretability analyses suggested its highlighted regions were consistent with known response-related pathology.

The authors suggest the approach could support risk stratification and future response-adapted clinical studies, but they frame it as a candidate rather than a validated clinical tool.

Sources

  1. Peer-reviewedLaboratory Investigation2026-10-09

The debate