Multiscale-fusion models with genomic, topological and pathomic features to predict response to radiation therapy for non-small cell lung cancer patients.
Journal:
Laboratory investigation; a journal of technical methods and pathology
Published Date:
Jun 12, 2025
Abstract
Artificial intelligence models with biomarkers to predict treatment responses to radiation would be necessary to maximise the treatment outcomes of individual patients, especially with histopathology images routinely obtained before treatment. We hypothesised that multiscale features, like genomic, pathomic and topological features, could be associated with the radiation response. We investigated fusion models with multiscale features in histopathology images to predict response to radiation therapy for patients (responders) with non-small cell lung cancer. Ten radiosensitivity-related (radiosensitive and radioresistant) genes were deployed as genomic features. Pathomic features were extracted from histopathology images by conventional pathomic analyses. Topological features represent the intrinsic properties of tumour cells using Betti numbers, which are mathematical invariants. We analysed non-small cell lung cancer (NSCLC) patients from TCGA and CPTAC who received radiotherapy and established three base models with genomic, topological, and pathomic features, respectively, and three fusion models. The topological model showed a higher area under the receiver operating characteristic curve (AUC) of 0.707 (p-value=0.026, log-rank test in overall survival analysis) in the internal test dataset and 0.720 (p-value=0.136) in the external test dataset. The results indicated that the topological models achieved better classification and prognostic prediction powers than the other base models. The inner-cell topological structure may have the ability to reveal the cell radiosensitivity-related information. Furthermore, the best fusion model with genomic, topological, and pathomic features achieved the highest AUC of 0.846 (p-value=0.019) and 0.731 (p-value=0.043) in predicting the treatment response and prognoses in the internal and external test datasets, respectively. This study demonstrated the predictive power of the multiscale fusion model for histopathology images, which may assist clinical physicians in the selection of responders to radiation for personalised radiation therapy and would be substantially beneficial for cancer patients.
Authors
Keywords
No keywords available for this article.
