Radiofrequency ablation (RFA) has been widely used as an alternative treatment modality for liver tumors. Monitoring the temperature distribution in the tissue during RFA is required to assess the thermal dosage. Ultrasound temperature imaging based on the detection of echo time shifts has received the most attention in the past decade. The coefficient k, connecting the temperature change and the echo time shift, is a medium-dependent parameter used to describe the confounding effects of changes in the speed of sound and thermal expansion as temperature increases. The current algorithm of temperature estimate based on echo time shift detection typically uses a constant k, resulting in estimation errors when ablation temperatures are higher than 50°C. This study proposes an adaptive-k algorithm that enables the automatic adjustment of the coefficient k during ultrasound temperature monitoring of RFA. To verify the proposed algorithm, RFA experiments on in vitro porcine liver samples (total n = 15) were performed using ablation powers of 10, 15, and 20 W. During RFA, a clinical ultrasound system equipped with a 7.5-MHz linear transducer was used to collect backscattered signals for ultrasound temperature imaging using the constant- and adaptive-k algorithms. Concurrently, an infrared imaging system and thermocouples were used to measure surface temperature distribution of the sample and internal ablation temperatures for comparisons with ultrasound estimates. Experimental results demonstrated that the proposed adaptive-k method improved the performance in visualizing the temperature distribution. In particular, the estimation errors were also reduced even when the temperature of the tissue is higher than 50°C. The proposed adaptive-k ultrasound temperature imaging strategy has potential to serve as a thermal dosage evaluation tool for monitoring high-temperature RFA.