Thermoacoustic imaging (TAI) has the potential for detecting hemorrhagic stroke. However, in transcranial TAI, the speed of sound (SoS) between the skull and brain tissue varies significantly. Therefore, if the image reconstruction assumes a uniform SoS, accurately locating the hemorrhagic lesion becomes challenging. In this Letter, we propose a fast inhomogeneous layer back-projection (BP) method based on the basic boundary line with a statistical approach to reconstruct TA images for noninvasive and non-ionizing hemorrhage detection. To validate our proposed method, we conducted numerical simulations using real human skull data and two phantom transcranial TAI experiments. In the numerical simulation, the proposed method improves the structural similarity index measure from 0.034 879 for BP with uniform SoS to 0.624 44. The phantom experimental results demonstrate that the proposed method renders the targets in the reconstructed image more consistent with the real targets. In the case of considering a three-layer SoS distribution, the time reversal method requires 1 min and 37.391 s to reconstruct a 201 × 201 pixels TA image. Meanwhile, the proposed method accomplishes the same-sized TA image reconstruction in only 2.113 397 s. The simulation and experimental results indicate that the proposed method enhances TAI's ability for accurate and fast identification of cerebral hemorrhage.