Channelized Hotelling Observers and their application in measuring human performance

Abstract

The main goal of this work is to demonstrate how a Channelized Hotelling Observer works, to show its application and value in the field of medical physics and also, to highlight its advantages over other observers. Channelized Hotelling Observers are a variation of the Model Observers, which are applicable with 3D tomography for detection of masses. During image quality assessment, the classifier needs to make a decision concerning that specific image. This decision can be most likely one of two possible cases – “signal absent” and “signal present”. These decisions represent the result, generated as a response to the input image. For this purpose, the widely used 4-Alternative Four Choice (4AFC) algorithm is applied, where one trial consists of a batch of four images – one from “signal present” and three more from “signal absent”. If the system has correctly classified the image as “present”, this is counted as a “hit”, otherwise it is a “miss”. This way, the final success rate of the observer is computed as the number of hits over the number of trials. These observers also have a channel mechanism, which reduces the dimensionality of the explored images. However, there is a need to be a tuning phase for the channels, with regard to the task. The next phase of the implementation is training. Here, an expected signal is compared to the covariance of the images, which results in a signal template. Due to the lack of a sufficient number of images, often they are used in both training and evaluation phases. The final “reading” phase (evaluation) involves two main features – observer reproducibility for various lesions and dose level estimation. Acknowledgements This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 692097.