This paper presents a novel algorithm, aimed at automatic endocardial boundary (inner boundary) detection in myocardial opacification scenarios. The data acquisition protocol uses (on purpose) low mechanical index imaging (i.e., weak ultrasound signal), so that the acquired images are characterized by low signal-to-noise ratios. The proposed algorithm is based on converting the frames, given in Cartesian coordinates, into polar coordinates, and applying a set of filters in order to compute the initial estimation of the endocardial boundary. The final estimation of the endocardial boundary is produced by an error correction process, which uses both spatial and temporal filtering. The estimated boundaries are converted into Cartesian coordinates, for display. Our algorithm has been tested on nine cine-loops. The resulting myocardial outlines have been separately assessed by two clinicians, scoring each segment in each cine-loop on a scale between 5 (excellent) and 1 (completely unacceptable). The mean overall score is 3.8 +/- 0.8, which seems adequate. The same clinicians have also manually drawn the contours of the endocardial boundary for the end-systolic and the end-diastolic frames of each cine-loop. The results show, that the mismatch between the automatically determined outlines and the manually drawn outlines is of the same order of magnitude as the interobserver variability. These results further support the validity of our method.