MV ‐ HCNN : A Multiview Hormonal Convolutional Neural Network for Noninvasive Hormonal State Assessment Using Breast Thermal Imaging

Monitoring hormonal changes during pregnancy, lactation, and menopause is essential for women's health diagnostics and treatment planning, yet no clinically established imaging modality exists for routine, noninvasive hormonal monitoring. Hormonal fluctuations influence peripheral vasodilation and thermoregulation, potentially generating measurable skin temperature patterns. In this paper, we present MV‐HCNN (Multiview Hormonal Convolutional Neural Network), a novel multiview infrared imaging and deep learning system for noninvasive classification of hormonal states in women. The system integrates a protocol‐compliant long‐wave infrared (LWIR) camera, a standardized five‐view breast acquisition protocol, and a multiview CNN that fuses view‐specific features with cross‐view global correlations for robust classification. MV‐HCNN is trained and validated on thermal images from 105 376 women and evaluated on an independent dataset of 47 220 women in a prospective‐like setting. Hormonal states (menopause, pregnancy, lactation) were self‐reported via structured interviews, and age‐matched analyses were performed to reduce confounding. The proposed MV‐HCNN achieved area under the ROC curves (AUCs) of 0.98, 0.91, and 0.91 for pregnancy, lactation, and menopause, respectively, outperforming a radiomics‐based baseline across all tasks. Performance remained stable in age‐matched groups except for menopause, which showed a moderate decline (AUC = 0.77) due to physiological variability during the menopausal transition. This study provides large‐scale evidence that MV‐HCNN, a portable, radiation‐free, multiview LWIR imaging system leveraging both view‐specific features and cross‐view global correlations, can differentiate physiological hormonal states, enabling scalable, noninvasive longitudinal monitoring for women's health.