Quantitative conductivity reconstruction network for brain hemorrhage detection in thermoacoustic imaging

Abstract Background In recent years, brain hemorrhage has become one of the leading causes of death and disability worldwide. Microwave‐induced thermoacoustic imaging (MITAI) has demonstrated significant potential for detecting brain hemorrhages. Purpose However, qualitative thermoacoustic imaging alone is insufficient for characterizing physiological changes in hemorrhage areas, making it difficult to grade the extent of hemorrhage. Therefore, the purpose of this paper is to quantitatively reconstruct the electrical conductivity parameters of brain hemorrhage. Methods Given the pronounced changes in dielectric properties in the hemorrhage area, we focus on conductivity as the imaging target and have proposed a quantitative conductivity reconstruction network (QCR‐Net). This network achieves the end‐to‐end quantitative reconstruction from thermoacoustic signals to conductivity parameter firstly and is applied to brain hemorrhage detection. Due to the complex end‐to‐end mapping relationship, we decompose the reconstruction problem into two sub‐tasks, including specific absorption rate (SAR) and conductivity reconstruction, to weaken the nonlinearity and ill‐posedness of the reconstruction problem. Results Additionally, we validated the effectiveness of this network through finite element modeling simulations and in phantom experiments with porcine brain tissue, providing a reliable physical metric for brain hemorrhage detection. We conducted multiple numerical simulations for brain hemorrhage, with results showing that the reconstruction errors for single and dual targets were limited to 4.7 and 5.9, respectively. Correspondingly, in experiments with pig brains and skull phantoms validated the effectiveness of this network, with reconstruction errors for single and dual targets confined to 7.0 and 12.7, respectively, which validates the feasibility thermoacoustic quantitative imaging in the transcranial hemorrhage detection. The network can distinguish hemorrhage target as small as 3 mm at a confidence probability 95. Conclusions The above results validate the feasibility of QCR‐Net for transcranial hemorrhage detection and highlight its potential as a reliable tool for quantitative thermoacoustic imaging in clinical applications. This work provides a foundation for further development of MITAI as a non‐invasive, high‐resolution imaging modality for brain hemorrhage diagnosis and grading.