In this talk, we discuss how computer vision can facilitate the interpretation of medical imaging data, or help making inferences based on models of such data. In order to illustrate this presentation, several applications of medical imaging measurements and modeling are discussed, focusing in areas such as the correction of imaging artifacts that may occlude visual information, tumor detection, modeling and measurement in different imaging modalities.
When interpreting medical imaging data with computer vision, usually we are trying to describe anatomic structures (or medical phenomena) using one or more images, and reconstruct some of its properties based on imaging data (like shape, texture or color). Actually, this is an ill-posed problem that humans can learn to solve effortlessly, but computer algorithms often are prone to errors. Nevertheless, in some cases computers can surpass humans and interpret medical images more accurately, given the proper choice of models, as we will show in this talk.
Reconstructing interesting properties of real world objects or phenomena from captured imaging data involves solving an inverse problem, in which we seek to recover some unknowns given insufficient information to specify a unique solution. Therefore, we disambiguate between possible solutions relying on models based on physics, mathematics or statistics. Modeling the real world in all its complexity still is an open problem. However, if we know the phenomenon or object of interest, we can construct detailed models using specialized techniques and domain specific representations, that are efficient at describing reliably the measurements (or obtaining measurements in some cases). In this talk, we briefly overview some challenging problems in computer vision for medical imaging and measurements, with illustrations and insights about model selection and model-based prediction. Some of the applications discussed in this talk are: modeling tumor shape and size, and making inferences about its future growth or shrinkage; modeling relevant details in the background of medical images to discriminate them from useless background noise; and modeling shading artifacts to minimize their influence when detecting and measuring skin lesions in standard camera images.
Medical images contain a wealth of information, which makes modeling of medical images a challenging task. Therefore, medical images often are segmented into multiple elementary parts, simplifying their representation and changing the image model into something that is more meaningful, or easier to analyze and measure (e.g. by describing the objects boundaries by lines or curves, or the image segments by their textures, colors, etc.). Nevertheless, these simpler image elements may be easy to perceive visually but difficult to describe. For example, the texture of a skin lesion may not have an identifiable texture element or a model known a priori, and regardless of that skin lesion detection must be accurate and precise. Segmentation of medical imaging data segmentation and analysis still is an open question, and some current directions are discussed in this talk.
Computer vision and modeling are interrelated. Modeling imaging measurements often involves errors, and estimating the expected error of a model can be important in applications (e.g. estimating a tumor size and its potential growth, or shrinkage, in response to a treatment). This issue can be approached by adapting machine learning and pattern recognition techniques to solve problems in medical imaging measurements. Typically, a model has tuning parameters, and these tuning parameters may change the model complexity. We wish to minimize modeling errors and the model complexity, in other words, to get the ‘big picture’ we often sacrifice some of the small details. For example, estimating tumor growth (or shrinkage) in response to treatment requires modeling the tumor shape and size, which can be challenging for real tumors, and simplified models may be justifiable if the predictions obtained are informative (e.g. to evaluate the treatment effectiveness). To conclude this talk, we outline the current trends in computer vision in medical imaging measurements, and discuss some open problems.