MRI-based deep learning radiomics models predict lung cancer brain metastases’ responses to targeted therapy

Résumé fourni par la source

Abstract Background Brain metastases (BrMs) are the most common intracranial neoplasms, with lung cancer serving as the predominant primary source. Targeted therapy offers a powerful treatment option for patients with BrMs. It is critical to predict the response of BrMs to targeted therapy prior to treatment to screen out patients who may benefit from it. The purpose of this study is to develop and validate MRI-based deep learning radiomics models (DLRMs) for predicting BrMs responses to targeted therapy in lung cancer patients. Methods 765 BrMs from 151 lung cancer patients who received targeted therapy were retrospectively included from seven centers. 467 BrMs were assigned to the training cohort, 192 BrMs to the internal validation cohort, and 106 BrMs to the external test set. Follow-up brain MRIs were used to assess each BrM’s response status. Handcrafted and deep learning (DL) signatures were constructed from pretreatment BrM MR images using the LASSO method, respectively. Two DLRMs were established by integrating the handcrafted and DL signatures based on the LASSO logistic regression coefficients to predict the BrM 6-month and 12-month responses to targeted therapy, respectively. DLRMs’ performance was evaluated by the area under curves (AUCs) and compared with handcrafted or DL signatures by the DeLong test. Results The AUCs of DLRM in predicting BrM 6-month response to targeted therapy were 0.848, 0.801, and 0.790 in the training, internal validation, and external test cohorts, respectively. The AUCs of DLRM in predicting BrM 12-month response to targeted therapy were 0.900, 0.818, and 0.782 in the training, internal validation, and external test cohorts, respectively. DLRMs outperformed handcrafted and DL signatures in predicting targeted therapy responses across all cohorts (all p < 0.05). Decision curve analysis showed that the DLRMs could benefit lung cancer patients with BrMs. Conclusion MRI-based DLRMs could predict BrM responses to targeted therapy across 6- and 12-month periods, which can assist in optimizing treatment for lung cancer patients who suffer from BrM.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.