Summary
This study addresses a fundamental problem in breast cancer management: patients with similar pathological diagnoses often show markedly different disease courses and treatment responses. To resolve this heterogeneity, the authors developed a multimodal AI framework that integrates pathology whole-slide images, lncRNA expression profiles, immune-cell infiltration scores, and clinical data to identify latent immune–metabolic tumor states. lncRNAs (long noncoding RNAs) are regulatory RNA molecules that influence immune activity and metabolic processes within the tumor microenvironment.
The scientific novelty lies in the identification of four immune–metabolic subtypes that are not captured by conventional molecular or pathological classifications. These subtypes showed clear and statistically significant separation in disease-free and overall survival (log-rank p < 0.001), demonstrating that immune activity must be interpreted together with metabolic context to explain patient outcomes. Importantly, the study showed that immune-cell abundance alone was insufficient; metabolic state determined whether immune infiltration translated into effective antitumor responses.
To support clinical translation, the authors developed an AI-based pathology model (DeepClinMed-IM) that predicts immune–metabolic subtypes directly from routine histopathology slides. In the training cohort, the model achieved strong discrimination with AUCs ranging from 0.89 to 0.93. In the independent validation cohort, performance remained robust though more variable (AUCs of 0.84, 0.78, 0.87, and 0.86 across subtypes), reflecting real-world heterogeneity, particularly for the immune–amino acid subtype. Notably, the multimodal prognostic model (DeepClinMed-PGM), which integrates pathology with molecular and clinical features, achieved higher and more stable performance, with external testing AUCs reaching approximately 0.90–0.93.
The impact of this work is its reframing of breast cancer heterogeneity as a coupled immune–metabolic phenomenon rather than a purely genetic one. By combining scalable pathology-based inference with multimodal prognostic modeling, the study provides a clinically feasible pathway for improving risk stratification and guiding immunotherapy decisions, even in settings where comprehensive molecular testing is limited.
Y. Yu, G. Cai, R. Lin, et al., “Multimodal data fusion AI model uncovers tumor microenvironment immunotyping heterogeneity and enhanced risk stratification of breast cancer,” MedComm, vol. 5, no. 12, Art. no. e70023, Dec. 2024, doi: 10.1002/mco2.70023
