Our analysis of trend topics supported the relevance of autism spectrum disorders as a central theme and identified “deep learning” as the niche theme that is mostly reported in recent articles. This finding might be explained by the fact that, in the last few years, a number of studies have started to explore the utility of eye tracking markers to improve the diagnosis of autism spectrum disorders. While different pipelines to analyze eye tracking data are available, increasing attention is now focused on the development of visual attention models using machine learning methods
[32][33][34][35][36][46,47,48,49,50]. Since the diagnosis of autism is challenging and no biomarker is available
[37][51], the development of computational models based on early abnormalities such as the differences in gaze processing might be of substantial help to improve and anticipate the diagnosis, thus, making it possible to initiate treatment at an earlier stage, when it is most effective
[38][52]. Eye tracking measurements that might prove to be useful as early biomarkers include dysregulations in pupil dilation
[39][40][41][53,54,55], changes in saccadic behavior, differences in gaze patterns during vision exposure to social stimuli
[42][43][44][56,57,58] and analysis of scan paths or gaze patterns
[45][46][47][48][49][50][59,60,61,62,63,64]. Some studies combined eye tracking data with other measurements such as resting-state EEG data
[51][52][53][65,66,67]. Using this approach, Kang and colleagues showed a classification accuracy of up to 85.44% using a support vector machine (SVM) classifier to discriminate children aged from 3 to 6 with autism from controls
[51][65]. Even higher performances were shown by another study conducted by Li and colleagues, using the three-layer Long Short-Term Memory (LSTM) network to discriminate 136 children with autism compared with 136 typically developing children based on gaze patterns during observations of 272 videos
[54][68]. In this study, accuracy improved from 86.4% to 92.6% using LSTM compared to SVM
[54][68]. Overall, results from these studies appear to be promising, although further research will be needed to evaluate their potential utility in the clinical setting and better assess the role of potential confounding factors that might affect gaze patterns (such as age, disease severity and duration, cognitive functioning and so on). Similarly, eye tracking measurements might be useful as prognostic markers
[55][69] or to evaluate the efficacy of different types of interventions
[56][57][58][59][70,71,72,73].
While we conducted a comprehensive analysis with no date and language restrictions, findings from our article must be interpreted in light of some limitations. First, our search might have missed some articles in which the terms included in our search strategy were mentioned in the main text but not in the title, abstract or keywords. In addition, our search might include some articles that mentioned the terms included in our search strategy but in which eye tracking experiments were not conducted (for instance reviews incorrectly classified by WOS as research articles, or articles in which new analytical methods were proposed). Despite these limitations, our analysis of large numbers of documents provides an updated and comprehensive picture of the use of eye tracking to study different medical conditions and highlights relevant trends with regards to investigated disorders and specific applications.
In conclusion, the eye tracking technique is increasingly being used in the study of different medical conditions. In autism spectrum disorders, eye tracking is widely used to evaluate eye movement abnormalities during face processing or looking at other social stimuli. Studies developing machine learning models provided promising evidence to support the utility of eye tracking measurements as biomarkers to discriminate patients with autism spectrum disorders from typically developing controls.