Abstract: Evaluation of the performance and limitations of medical imaging algorithms is essential to estimate their impact in social, economic or clinical aspects. However, validation of medical imaging techniques is a challenging task due to the variety of imaging and clinical problems involved, as well as, the difficulties for systematically extracting a reliable solely ground truth. Although specific validation protocols are reported in any medical imaging paper, there are still two major concerns: definition of standardized methodologies transversal to all problems and generalization of conclusions to the whole clinical data set.
We claim that both issues would be fully solved if we had a statistical model relating ground truth and the output of computational imaging techniques. Such a statistical model could conclude to what extent the algorithm behaves like the ground truth from the analysis of a sampling of the validation data set. We present a statistical inference framework reporting the agreement and describing the relationship of two quantities. We show its transversality by applying it to validation of two different tasks: contour segmentation and landmark correspondence.

Abstract: A main challenge in Evolutionary Algorithms (EAs) is determining a termination condition ensuring stabilization close to the optimum in real-world applications. Although for known test functions distribution-based quantities are good candidates (as far as suitable parameters are used), in real-world problems an open question still remains unsolved. How can we estimate an upper-bound for the termination condition value ensuring a given accuracy for the (unknown) EA solution?
We claim that the termination problem would be fully solved if we defined a quantity (depending only on the EA output) behaving like the solution accuracy. The open question would be, then, satisfactorily answered if we had a model relating both quantities, since accuracy could be predicted from the alternative quantity. We present a statistical inference framework addressing two topics: checking the correlation between the two quantities and defining a regression model for predicting (at a given confidence level) accuracy values from the EA output.

Abstract: Optical flow is a valuable tool for motion analysis in medical imaging sequences. A reliable application requires determining the accuracy of the computed optical flow. This is a main challenge given the absence of ground truth in medical sequences. This paper presents an error analysis of Lucas-Kanade schemes in terms of intrinsic design errors and numerical stability of the algorithm. Our analysis provides a confidence measure that is naturally correlated to the accuracy of the flow field. Our experiments show the higher predictive value of our confidence measure compared to existing measures.

Abstract: Assessing Optical Flow (OF) quality is essential for its further use in reliable decision support systems. The absence of ground truth in such situations leads to the computation of OF Confidence Measures (CM) obtained from either input or output data. A fair comparison across the capabilities of the different CM for bounding OF error is required in order to choose the best OF-CM pair for discarding points where OF computation is not reliable. This paper presents a statistical probabilistic framework for assessing the quality of a given CM. Our quality measure is given in terms of the percentage of pixels whose OF error bound can not be determined by CM values. We also provide statistical tools for the computation of CM values that ensures a given accuracy of the flow field.