Objective This study aimed to determine whether quantitative volumetric measures from MRI and metabolic indices from FDG-PET can independently or jointly predict surgical outcome in patients with drug-resistant temporal lobe epilepsy (TLE), particularly when accounting for the well-established prognostic role of hippocampal sclerosis (HS) co-pathology. Methods We retrospectively reviewed patients who underwent temporal lobe resection for drug-resistant epilepsy, comprising 80 with MRI data, 77 with FDG-PET data, and a subset of 42 with both modalities available (fusion cohort). Quantitative asymmetry indices for temporal lobe subregions were derived using the AAL atlas. To assess whether imaging features contributed prognostic information beyond HS status and clinical covariates, we employed hierarchical logistic regression. Data-driven predictive performance was further evaluated through machine learning models using nested leave-one-out cross-validation with permutation testing. Results Across all cohorts, HS co-pathology consistently emerged as the strongest predictor of favorable outcome (MRI cohort: OR=18.4, P<0.001; PET cohort: OR=42.0, P<0.001). When examined individually, neither MRI-derived nor PET-derived quantitative features added significantly to the predictive model beyond HS (MRI: P=0.085; PET: P=0.386). By contrast, combining both modalities in the fusion cohort yielded a significant incremental contribution over HS and clinical variables (likelihood ratio test, P=0.009), with the AUC rising from 0.778 to 0.963. Of particular interest, amygdala volumetric asymmetry on MRI was identified as an independent predictor not previously reported (OR=83.7, P=0.041). Machine learning approaches yielded only modest discrimination (fusion cohort AUC=0.692, P=0.075) and did not outperform the hypothesis-driven statistical framework. Conclusion Integrating MRI volumetric and FDG-PET metabolic data offers meaningful prognostic value that extends beyond what HS status alone can provide. Amygdala asymmetry on MRI represents a novel independent predictor warranting further validation. Our findings favor comprehensive multimodal presurgical workup over reliance on a single imaging modality and suggest that, for surgical outcome prediction in TLE, hypothesis-driven analytical approaches may hold advantages over purely data-driven machine learning strategies.
ObjectiveTo summarize the current research progress of imaging in evaluating the efficacy of neoadjuvant therapy for breast cancer, analyze the clinical application value and limitations of conventional imaging modalities, and explore the application prospects of advanced technologies such as multimodal fusion, radiomics, and artificial intelligence in efficacy prediction. MethodBy searching recent domestic and international literature on imaging evaluation of neoadjuvant therapy for breast cancer, this review summarizes traditional imaging evaluation methods and the applications of cutting-edge technologies such as multimodal fusion, radiomics, and deep learning. ResultsTraditional imaging methods play an important role in the evaluation of post-treatment efficacy, but there are inherent limitations. The morphology-based evaluation paradigm struggles to meet the clinical demand for early prediction and exhibits a significant gap compared with the histopathological gold standard. In recent years, the development of multimodal fusion, radiomics, and deep learning technologies has demonstrated promising efficacy in predicting pathologic complete response. However, most studies are single-center retrospective designs, and their external validation and interpretability still need to be strengthened. ConclusionsImaging enables both retrospective assessment and prospective prediction in evaluating the efficacy of neoadjuvant therapy for breast cancer. In the future, the establishment of standardized imaging acquisition and validation databases, the development of hybrid models that combine predictive performance with interpretability, and the promotion of the deep integration of imaging with multi-omics data will be important directions for clinical transformation.