Frontiers in endocrinology · 2026
Habitat-radiomics combining multichannel 2.5D deep learning for differentiating adrenal adenomas from metastases using automatic segmentation: a multicenter study.
СтатьяМногоцентровое исследованиеНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Yin Shengnan, Ding Ning, Yang Chuqi, Wang Shaocai, Li Mengjuan, Ji Yiding, Liu Tong, Jin Long
Аннотация
BACKGROUND: The qualitative diagnosis of lipid-poor adrenal adenomas and metastases presents challenges, yet there is a significant difference in their treatment principles and prognosis.
MATERIALS AND METHODS: A total of 390 patients from two hospitals were divided into training, internal validation, and external test sets. Lesion segmentation was performed automatically using the large segmentation model Medical SAM. The 2.5D deep learning model was constructed using the DenseNet-121 architecture. Habitat-radiomics employed the K-means clustering algorithm; both habitat-radiomics and conventional radiomics features were extracted from lesion regions using the PyRadiomics toolkit, with XGBoost machine learning models subsequently developed. The fusion model incorporated the 2.5D deep learning model scores, habitat-radiomics features, and clinical features.
RESULTS: The fusion model demonstrated the best overall performance, achieving areas under the ROC curve (AUC) of 0.983, 0.913, and 0.886 in the training, internal validation, and external test sets, respectively. The standalone 2.5D deep learning and habitat-radiomics models also showed good predictive performance, with AUCs ranging from 0.847-0.980, 0.805-0.967, respectively.
CONCLUSION: The fusion model holds potential for noninvasively differentiating lipid-poor adrenal adenomas from metastases and may provide a valuable decision-making basis for subsequent precision treatment.
MATERIALS AND METHODS: A total of 390 patients from two hospitals were divided into training, internal validation, and external test sets. Lesion segmentation was performed automatically using the large segmentation model Medical SAM. The 2.5D deep learning model was constructed using the DenseNet-121 architecture. Habitat-radiomics employed the K-means clustering algorithm; both habitat-radiomics and conventional radiomics features were extracted from lesion regions using the PyRadiomics toolkit, with XGBoost machine learning models subsequently developed. The fusion model incorporated the 2.5D deep learning model scores, habitat-radiomics features, and clinical features.
RESULTS: The fusion model demonstrated the best overall performance, achieving areas under the ROC curve (AUC) of 0.983, 0.913, and 0.886 in the training, internal validation, and external test sets, respectively. The standalone 2.5D deep learning and habitat-radiomics models also showed good predictive performance, with AUCs ranging from 0.847-0.980, 0.805-0.967, respectively.
CONCLUSION: The fusion model holds potential for noninvasively differentiating lipid-poor adrenal adenomas from metastases and may provide a valuable decision-making basis for subsequent precision treatment.