In this paper, the estimation of acoustic drug release from micelles is addressed. The release is measured as a decrease in fluorescence once ultrasound is applied. Initially, a Kalman filter is used to fuse the drug encapsulation (calculated as 100 %-release%) dynamics and measurements. Since the measurements' noise statistics are not known a priori, the encapsulation estimate is not optimal. Therefore, an approach is proposed to adaptively estimate the drug release given the statistical properties of the measurements. In this approach, a number of measurement covariance magnitudes are hypothesized. A Kalman filter is used to obtain the estimate of the acoustic release given each hypothesized measurement noise covariance. Simultaneously, the probabilities of these measurement covariance hypotheses are sequentially computed as the measurements and the predicted release estimates are obtained. Finally, the optimal release estimate is obtained by probabilistically adding the estimates from the hypothesized Kalman filter estimates. The proposed algorithms are first tested using a simulation environment. Subsequently, experimental results are shown to validate their performance. The experiments conducted cover various ultrasonic power densities for both non-targeted and targeted micelles.