Integrating handcrafted and deep learning MRI signatures: an interpretable framework for predicting chemotherapy benefit in glioma - PubMed
3 hours ago
- #multimodal MRI
- #AI in oncology
- #glioma chemotherapy
- The study presents a multimodal AI framework integrating handcrafted radiomics and deep learning features from MRI to predict postoperative chemotherapy benefit in glioma.
- Using data from the TCGA-LGG cohort, it extracted 726 radiomics features and generated deep learning embeddings via 3D CNN and ResNet-18 models.
- Prediction models were built using radiomics alone, deep learning alone, and their fusion, evaluated on real-world and augmented datasets.
- The fusion model achieved the best performance, with an AUC of 0.75 on Dataset 1 and 0.99 on Dataset 2, highlighting complementary information between feature types.
- Interpretability was assessed using SHAP and Grad-CAM, revealing key radiomics features and attention on tumor core and infiltrative margins relevant to drug efficacy.
- The study emphasizes data quality and feature diversity as core drivers for accurate drug response prediction in glioma treatment.