Deep learning has revolutionized thermographic imaging by enabling accurate analysis of complex datasets for breast tumor detection. However, the accurate estimation of breast tumor size from temperature features presents a critical challenge, due to the need for high-quality data acquisition systems, robust preprocessing techniques, and access to extensive labeled datasets. In this study, we deploy a deep learning model for automated tumor size prediction using an innovative wearable system based on flexible thermal micro-biosensors. Our system combines custom-designed hardware for continuous temperature monitoring with a Feed-Forward Deep Neural Network (FF-DNN) trained previously on simulated thermal data. By collecting the experimental data from a breast phantom, including different tumor cases, we evaluated the prediction performance of the pre-trained FF-DNN model in predicting tumor size from extracted temperature features. The results demonstrate the successful transfer learning from simulated to experimental thermographic data with a minimum error of 0.527 mm in estimating tumor diameter. This proposed hardware-software approach advances the field of smart wearable thermographic systems for accessible and non-invasive breast cancer diagnosis.