Evaluation of Gaze Tracking Calibration for Longitudinal Biomedical Imaging Studies

Assessing the data quality of AdHawk MindLink eye-tracking glasses

Most commercially available eye-tracking devices rely on video cameras and image processing algorithms to track gaze. Despite this, emerging technologies are entering the field, making high-speed, cameraless eye-tracking more accessible. In this study, a series of tests were conducted to compare the data quality of MEMS-based eye-tracking glasses (AdHawk MindLink) with three widely used camera-based eye-tracking devices (EyeLink Portable Duo, Tobii Pro Glasses 2, and SMI Eye Tracking Glasses 2). The data quality measures assessed in these tests included accuracy, precision, data loss, and system latency. The results suggest that, overall, the data quality of the eye-tracking glasses was lower compared to that of a desktop EyeLink Portable Duo eye-tracker. Among the eye-tracking glasses, the accuracy and precision of the MindLink eye-tracking glasses were either higher or on par with those of Tobii Pro Glasses 2 and SMI Eye Tracking Glasses 2. The system latency of MindLink was approximately 9 ms, significantly lower than that of camera-based eye-tracking devices found in VR goggles. These results suggest that the MindLink eye-tracking glasses show promise for research applications where high sampling rates and low latency are preferred.

The Use of Machine Learning in Eye Tracking Studies in Medical Imaging: A Review

Machine learning (ML) has revolutionized medical image-based diagnostics. In this review, we cover a rapidly emerging field that can be potentially significantly impacted by ML - eye tracking in medical imaging. The review investigates the clinical, algorithmic, and hardware properties of the existing studies. In particular, it evaluates 1) the type of eye-tracking equipment used and how the equipment aligns with study aims; 2) the software required to record and process eye-tracking data, which often requires user interface development, and controller command and voice recording; 3) the ML methodology utilized depending on the anatomy of interest, gaze data representation, and target clinical application. The review concludes with a summary of recommendations for future studies, and confirms that the inclusion of gaze data broadens the ML applicability in Radiology from computer-aided diagnosis (CAD) to gaze-based image annotation, physicians' error detection, fatigue recognition, and other areas of potentially high research and clinical impact.

Eye tracking: empirical foundations for a minimal reporting guideline

In this paper, we present a review of how the various aspects of any study using an eye tracker (such as the instrument, methodology, environment, participant, etc.) affect the quality of the recorded eye-tracking data and the obtained eye-movement and gaze measures. We take this review to represent the empirical foundation for reporting guidelines of any study involving an eye tracker. We compare this empirical foundation to five existing reporting guidelines and to a database of 207 published eye-tracking studies. We find that reporting guidelines vary substantially and do not match with actual reporting practices. We end by deriving a minimal, flexible reporting guideline based on empirical research (Section "An empirically based minimal reporting guideline").

Deeply Encoding Stable Patterns From Contaminated Data for Scenery Image Recognition

Effectively recognizing different sceneries with complex backgrounds and varied lighting conditions plays an important role in modern AI systems. Competitive performance has recently been achieved by the deep scene categorization models. However, these models implicitly hypothesize that the image-level labels are 100% correct, which is too restrictive. Practically, the image-level labels for massive-scale scenery sets are usually calculated by external predictors such as ImageNet-CN. These labels can easily become contaminated because no predictors are completely accurate. This article proposes a new deep architecture that calculates scene categories by hierarchically deriving stable templates, which are discovered using a generative model. Specifically, we first construct a semantic space by incorporating image-level labels using subspace embedding. Afterward, it is noticeable that in the semantic space, the superpixel distributions from identically labeled images remain unchanged, regardless of the image-level label noises. On the basis of this observation, a probabilistic generative model learns the stable templates for each scene category. To deeply represent each scenery category, a novel aggregation network is developed to statistically concatenate the CNN features learned from scene annotations predicted by HSA. Finally, the learned deep representations are integrated into an image kernel, which is subsequently incorporated into a multiclass SVM for distinguishing scene categories. Thorough experiments have shown the performance of our method. As a byproduct, an empirical study of 33 SIFT-flow categories shows that the learned stable templates remain almost unchanged under a nearly 36% image label contamination rate.

First Trimester Gaze Pattern Estimation Using Stochastic Augmentation Policy Search for Single Frame Saliency Prediction

While performing an ultrasound (US) scan, sonographers direct their gaze at regions of interest to verify that the correct plane is acquired and to interpret the acquisition frame. Predicting sonographer gaze on US videos is useful for identification of spatio-temporal patterns that are important for US scanning. This paper investigates utilizing sonographer gaze, in the form of gaze-tracking data, in a multimodal imaging deep learning framework to assist the analysis of the first trimester fetal ultrasound scan. Specifically, we propose an encoderdecoder convolutional neural network with skip connections to predict the visual gaze for each frame using 115 first trimester ultrasound videos; 29,250 video frames for training, 7,290 for validation and 9,126 for testing. We find that the dataset of our size benefits from automated data augmentation, which in turn, alleviates model overfitting and reduces structural variation imbalance of US anatomical views between the training and test datasets. Specifically, we employ a stochastic augmentation policy search method to improve segmentation performance. Using the learnt policies, our models outperform the baseline: KLD, SIM, NSS and CC (2.16, 0.27, 4.34 and 0.39 versus 3.17, 0.21, 2.92 and 0.28).

Transforming obstetric ultrasound into data science using eye tracking, voice recording, transducer motion and ultrasound video

Ultrasound is the primary modality for obstetric imaging and is highly sonographer dependent. Long training period, insufficient recruitment and poor retention of sonographers are among the global challenges in the expansion of ultrasound use. For the past several decades, technical advancements in clinical obstetric ultrasound scanning have largely concerned improving image quality and processing speed. By contrast, sonographers have been acquiring ultrasound images in a similar fashion for several decades. The PULSE (Perception Ultrasound by Learning Sonographer Experience) project is an interdisciplinary multi-modal imaging study aiming to offer clinical sonography insights and transform the process of obstetric ultrasound acquisition and image analysis by applying deep learning to large-scale multi-modal clinical data. A key novelty of the study is that we record full-length ultrasound video with concurrent tracking of the sonographer's eyes, voice and the transducer while performing routine obstetric scans on pregnant women. We provide a detailed description of the novel acquisition system and illustrate how our data can be used to describe clinical ultrasound. Being able to measure different sonographer actions or model tasks will lead to a better understanding of several topics including how to effectively train new sonographers, monitor the learning progress, and enhance the scanning workflow of experts.

Machine learning-based analysis of operator pupillary response to assess cognitive workload in clinical ultrasound imaging

Introduction

Pupillometry, the measurement of eye pupil diameter, is a well-established and objective modality correlated with cognitive workload. In this paper, we analyse the pupillary response of ultrasound imaging operators to assess their cognitive workload, captured while they undertake routine fetal ultrasound examinations. Our experiments and analysis are performed on real-world datasets obtained using remote eye-tracking under natural clinical environmental conditions.

Methods

Our analysis pipeline involves careful temporal sequence (time-series) extraction by retrospectively matching the pupil diameter data with tasks captured in the corresponding ultrasound scan video in a multi-modal data acquisition setup. This is followed by the pupil diameter pre-processing and the calculation of pupillary response sequences. Exploratory statistical analysis of the operator pupillary responses and comparisons of the distributions between ultrasonographic tasks (fetal heart versus fetal brain) and operator expertise (newly-qualified versus experienced operators) are performed. Machine learning is explored to automatically classify the temporal sequences into the corresponding ultrasonographic tasks and operator experience using temporal, spectral, and time-frequency features with classical (shallow) models, and convolutional neural networks as deep learning models.

Results

Preliminary statistical analysis of the extracted pupillary response shows a significant variation for different ultrasonographic tasks and operator expertise, suggesting different extents of cognitive workload in each case, as measured by pupillometry. The best-performing machine learning models achieve receiver operating characteristic (ROC) area under curve (AUC) values of 0.98 and 0.80, for ultrasonographic task classification and operator experience classification, respectively.

Conclusion

We conclude that we can successfully assess cognitive workload from pupil diameter changes measured while ultrasound operators perform routine scans. The machine learning allows the discrimination of the undertaken ultrasonographic tasks and scanning expertise using the pupillary response sequences as an index of the operators’ cognitive workload. A high cognitive workload can reduce operator efficiency and constrain their decision-making, hence, the ability to objectively assess cognitive workload is a first step towards understanding these effects on operator performance in biomedical applications such as medical imaging.

•

Machine learning-based pupillary response analysis is performed to assess operator cognitive workload in clinical ultrasound.

•

A systematic multi-modal data analysis pipeline is proposed using eye-tracking, pupillometry, and sonography data science.

•

Pertinent challenges of natural or real-world clinical datasets are addressed.

•

Pupillary responses around event triggers, different ultrasonographic tasks, and different operator experiences are studied.

•

Machine learning models are learnt to classify undertaken tasks or operator expertise from pupillometric time-series data.

Multi-Modal Learning from Video, Eye Tracking, and Pupillometry for Operator Skill Characterization in Clinical Fetal Ultrasound

This paper presents a novel multi-modal learning approach for automated skill characterization of obstetric ultrasound operators using heterogeneous spatio-temporal sensory cues, namely, scan video, eye-tracking data, and pupillometric data, acquired in the clinical environment. We address pertinent challenges such as combining heterogeneous, small-scale and variable-length sequential datasets, to learn deep convolutional neural networks in real-world scenarios. We propose spatial encoding for multi-modal analysis using sonography standard plane images, spatial gaze maps, gaze trajectory images, and pupillary response images. We present and compare five multi-modal learning network architectures using late, intermediate, hybrid, and tensor fusion. We build models for the Heart and the Brain scanning tasks, and performance evaluation suggests that multi-modal learning networks outperform uni-modal networks, with the best-performing model achieving accuracies of 82.4% (Brain task) and 76.4% (Heart task) for the operator skill classification problem.

Expected-value bias in routine third-trimester growth scans

OBJECTIVES

Operators performing fetal growth scans are usually aware of the gestational age of the pregnancy, which may lead to expected-value bias when performing biometric measurements. We aimed to evaluate the incidence of expected-value bias in routine fetal growth scans and assess its impact on standard biometric measurements.

METHODS

We collected prospectively full-length video recordings of routine ultrasound growth scans coupled with operator eye tracking. Expected value was defined as the gestational age at the time of the scan, based on the estimated due date that was established at the dating scan. Expected-value bias was defined as occurring when the operator looked at the measurement box on the screen during the process of caliper adjustment before saving a measurement. We studied the three standard biometric planes on which measurements of head circumference (HC), abdominal circumference (AC) and femur length (FL) are obtained. We evaluated the incidence of expected-value bias and quantified the impact of biased measurements.

RESULTS

We analyzed 272 third-trimester growth scans, performed by 16 operators, during which a total of 1409 measurements (354 HC, 703 AC and 352 FL; including repeat measurements) were obtained. Expected-value bias occurred in 91.4% of the saved standard biometric plane measurements (85.0% for HC, 92.9% for AC and 94.9% for FL). The operators were more likely to adjust the measurements towards the expected value than away from it (47.7% vs 19.7% of measurements; P < 0.001). On average, measurements were corrected by 2.3 ± 5.6, 2.4 ± 10.4 and 3.2 ± 10.4 days of gestation towards the expected gestational age for the HC, AC, and FL measurements, respectively. Additionally, we noted a statistically significant reduction in measurement variance once the operator was biased (P = 0.026). Comparing the lowest and highest possible estimated fetal weight (using the smallest and largest biased HC, AC and FL measurements), we noted that the discordance, in percentage terms, was 10.1% ± 6.5%, and that in 17% (95% CI, 12-21%) of the scans, the fetus could be considered as small-for-gestational age or appropriate-for-gestational age if using the smallest or largest possible measurements, respectively. Similarly, in 13% (95% CI, 9-16%) of scans, the fetus could be considered as large-for-gestational age or appropriate-for-gestational age if using the largest or smallest possible measurements, respectively.

CONCLUSIONS

During routine third-trimester growth scans, expected-value bias frequently occurs and significantly changes standard biometric measurements obtained. © 2019 the Authors. Ultrasound in Obstetrics & Gynecology published by John Wiley & Sons Ltd on behalf of the International Society of Ultrasound in Obstetrics and Gynecology.