The participants (all males) for this study were recruited via advertisements and announcements in driving schools and social media. Initially, 81 subjects (39 novice and 42 experienced drivers) agreed to participate after the procedure of the study was explained to them. After the first screening that included driving experience and accident record (novice drivers: less than 6 months period of driving license acquisition and without involvement in any fatal accident; experienced drivers: more than 10 years of driving experience and without any fatal accident and have driving as a daily activity), negative history of neurological and psychological disorders or major systematic illness such as migraine headaches, diabetes, heart disease, head injury, hospitalization for head surgery/illness and absence of non-removable metal in the whole body or metallic tattoos around the head, smoking and addiction to drug/alcohol, 19 novice and 25 experienced drivers were included.

After their arrival at the National Brain Mapping Laboratory for fMRI investigation, secondary screening that included depression, anxiety, stress score (DASS 21), Barratt impulsiveness score (BIS 11), Pittsburgh sleep quality index (PSQI) questionnaires and clinical examination was carried out. Considering the effect of depression, anxiety, stress, impulsiveness and sleep quality on visual task performance (including HPT), all the participants were evaluated with validated Persian version of DASS 21, BIS and PSQI written questionnaires (Asghari-Moghadam et al., 2010[2]; Mokri et al., 2008[49]; Nazifi et al., 2014[51]). Those who had high depression, anxiety and stress score (DASS) or high impulsiveness (low BIS) or low sleep quality were excluded from the study. The criteria included in the study were as follows:

Depression score (from DASS 21) under 9

Based on the results of depression, anxiety and stress score (mean ± SD = 9.91 ± 6.18), Barratt impulsiveness index (mean ± SD = 92.66 ± 10.61) and Pittsburgh Sleep Quality Index (mean ± SD = 3.88 ± 1.6, sleep quality = 1, sleep quantity = 6.65 ± 1.4), subjects were found to be in psychologically eligible state. In addition, all the subjects expressed in self-reported questionnaire that they played computer games less than two hours a week.

Finally, 34 healthy drivers (17 novice, mean age 22.29 ± 3.35, range 19-31 years; and 17 experienced drivers, mean age 42.53, ± 7.22, range 31-52 years), all right-handed, were included in this study for data collection. Two novice and 1 experienced drivers were excluded later from the data analyses because of the artifacts or data recording problems. After drop out, 15 novice drivers (mean age = 22.13 ± 2.38, range 19-27 years) and 16 experienced drivers (mean age = 41.44 ± 5.83, range 33-52 years) were included in fMRI analysis.

None of the participants had any medical restrictions or vision problems for undergoing MRI. Before fMRI, all the participants underwent complete physical and neurological examination. Screening and pre-test clinical examinations were conducted to determine whether the two groups of drivers were healthy and had no medical problems. All were found medically fit to undergo fMRI study.

The study was approved by the IRCCS TUMS Ethics Committee, and all the participants signed written informed consent (Ethics code: IR.TUMS.SPH.REC.1396.4053).

Figure 1(Fig. 1) (part A) shows the design of the study comprising of 5 sections. After screening tests, written informed consent and clinical examinations, subjects were familiarized with the test procedure at the National Brain Mapping Laboratory, University of Tehran, I.R. Iran. The subjects were verbally informed that in the scanner, clips of various traffic situations (urban, rural, interurban and suburb roads) are going to be shown and that they will see the scene from the front window of the camera car with driver’s view. They were informed of a probable developing hazard in each clip. The subjects were guided to press a button in their right hand whenever they perceive an approaching hazard that needs the driver to respond. They were also told that there were no collisions or questions in the scene. Figure 1(Fig. 1) (part B) shows the fMRI experiment task design. The experiment included 2 sections: Hazard perception task (400 sec.) and resting state (400 sec.). Stimuli were back projected. The watching time and sequence of videos from the beginning to end were same for all the subjects. The HPT was an event design played as a movie without interruptions, and it was same for all the subjects. In the resting state, all the subjects just watched a black screen for 400 sec. The total duration of whole experiment was 2 hours for each subject.

Based on the scenarios leading to accidents with extensive damages that occurred in various (urban, rural, interurban and suburb) locations, hazard perception video clips were prepared from various streets in urban Tehran and its suburbs with the help of police and experts using the validity criteria of Marrington et al. (2008[47]) and Wetton et al. (2010[81]). Out of the total 150 clips, 35 selected clips (8 ± 1 second) were arranged randomly, including hazard situations:

Cyclist behavior

Each clip had a specific (different from others) situation such as scene, hazard window and onset time. Between each hazard situation clip, a neutral scene (3 seconds) of a natural scenery was added. The total duration of this HPT was 400 sec. The subjects were informed before that the HPT (Figure 1(Fig. 1)) would begin with a neutral scene and in between hazard situation clips, the same neutral scene would be repeated signaling the end of clip. Sample video clips SV1 and SV2 are added as supplementary data files.

After HTP, subjects were given resting state for 400 seconds. During the resting state, a black screen was shown to the subjects. This imaging session was designed to detect differences of default mode network and perform Independent Component Analysis (ICA) between the two groups.

Brain imaging was conducted using 3T MR scanner (Siemens, Prisma model, Iran National Brain Mapping Center, Tehran, Iran) equipped with a 64-channel head coil. Functional images were acquired with T₂* –weighted echo-planar imaging (EPI); TR = 3 s, TE = 30 ms, flip angle = 90º, matrix 64 _ 64, voxel size = 3 _ 3 mm in-plane, slice thickness = 3 mm; 50 % distance factor; FOV = 448 mm, gap = 0.53 ms, 42 slices, covering the whole cerebral cortex. Resting state data were acquired with T₂* –weighted EPI (TR = 2500 ms, TE = 30 ms, flip angle = 90º, 42 axial slices, 3.0_3.0_3.0 mm³, distance factor = 25 %, FOV = 448_448 mm², interleaved slice ordering). Structural T₁ – weighted images were collected with a Magnetization-Prepared Rapid Gradient-Echo (MP-RAGE) sequence as an anatomical reference; TR = 2 s, TE = 3.53 ms, TI = 1100 ms, voxel size = 1 × 1 × 1 mm, matrix resolution 256 × 256 × 176, no gap, axial acquisition.

Results of ANOVA for the overall performance (mean Hazard Perception Reaction Times (HPRT) and mean Hazard Perception Miss Rates (HPMR)) in both the groups revealed no significant differences between driving situations with different hazard sources. For novice drivers: F _HPRT = 1.78, p _HPRT = 0.16, F _{miss rate} = 0.54, p _{miss rate} = 0.70 and degrees of freedom = 4, 30. For experienced drivers: F _HPRT = 0.72, p _HPRT = 0.58, F _{miss rate} = 0.50, p _{miss rate} = 0.73 and degrees of freedom = 4, 30. However, the overall performance of the experienced drivers compared to the novice ones in the hazard perception task was better.

The mean HPRT during the HPT in the MRI scanner were 3.58 ± 1.45 sec. and 1.32 ± 1.09 sec. for the novice drivers and experienced ones respectively. The t-test revealed significant effects of experience in HPRT. Significant difference between the two groups in HPRT (t_{2, 68} = 7.35, p < 0.001) showed that novice drivers had significantly longer HPRT compared to the experienced ones in all hazardous situations. Table 2(Tab. 2) shows the results of HPRT in the different hazard sources for the novice and experienced drivers. The results of ANOVA for the reaction times in both groups revealed no significant differences between different hazard sources (p > 0.05).

Table 2(Tab. 2) summarizes the miss rates of the two groups. The mean HPMR for the novice drivers were 39.67 ± 15.72 %, and 11.42 ± 8.36 % for the experienced drivers. The t-test for mean HPMR in the HPT that was performed inside the MRI topographer revealed a significant difference between the novice and the experienced drivers in all situations of road hazards (t_{2, 68}=8.99, p <0.001). Mean miss rate of HPT for the novice drivers was significantly (28.25 %) higher than that of the experienced drivers. In addition, the ANOVA for the miss rates in both the groups revealed no significant difference between different hazard sources (p > 0.05).

The HPT revealed significant activation of different areas of the brain (p< 0.05, corrected from multiple comparisons) associated with hazard perception [hit rate – without hit rate (including missing of hazard and neutral scene)] (Figure 2(Fig. 2)), including lateral occipital (x = 54, y = -56, z = 8); lingual gyrus (x = -2, y = -72, z = -10); insular cortex (x = 36, y =16, z = 0); FEF (x = 44, y = 2, z = 40) and cerebellum (x = -4, y = -72, z = -20) in all the subjects. This task was then examined in more detail by considering the mean activation within each group separately. In the majority of these areas, the results demonstrated higher brain activation in the experienced drivers compared to the novice ones.

Compared to the novice drivers, the experienced drivers showed greater brain activation during HPT in five areas. Figure 3(Fig. 3) shows the brain areas with significantly increased neural activity associated with HPT in the experienced group compared to the novice ones (red/yellow blobs) on MNI template. These included the occipital lobe (supracalcarine cortex (x = 0, y = -80, z = 8; Z = 4.68), lingual gyrus (x = 14, y = -58, z = -8; Z = 4.28)), bilateral SMG (x = -58, y = -48, z = 20; Z = 3.99 and x = +62, y = -42, z = 12; Z = 4.51), right anterior insular cortex (AIC) (x= 40, y= 12, z= -4; Z= 4.14), FEF (x= 0, y= 16, z= 46; Z= 3.82), ACC (x=0, y=32, z = 24; Z = 3.12) and cerebellum (x = -18, y = -76, z = -32; Z = 4) as shown in Table 3(Tab. 3).

The first ROI-to-ROI analysis was performed to investigate the brain regions that were engaged in hazard perception performance (visual, dorsal attention, salience networks of the brain) in the novice and experienced drivers. There were significant connectivity differences in the salience networks between the novice and experienced drivers.

The most significant connectivity differences between the two groups were related to the ACC (t = 9.34, p < 0.001). Figure 4(Fig. 4) shows the ACC connectivity with different regions in the two groups of drivers. The experienced drivers showed significantly stronger (positive) connectivity between ACC and right AIC (t = 5.24, p = 0.002) and between ACC and bilateral SMG (ACC-right SMG (t = 2.46; p = 0.041) and ACC-left SMG (t = 2.66; p = 0.03)). However, the experienced drivers did not show greater connectivity between ACC and bilateral RPFC or left anterior insula at the level of significance (i.e., p < 0.05, FDR corrected).

ROI-to-ROI analysis showed functional connectivity differences between the two groups in ACC and the regions of the salience network. Therefore, we carried out seed-to-voxel analysis to investigate whether there was greater connectivity with the regions outside the networks. Whole-brain between group seed-to-voxel analysis was performed to examine the potential group differences in the functional connectivity between ACC and all other voxels in the brain. The experienced group showed stronger (negative) connectivity between ACC and a large cluster of voxels in the IPS, bilateral occipital cortices and right FEF, as seen in Figure 4(Fig. 4). Thus, compared to the novice drivers, the experienced drivers showed different patterns of connectivity between the ACC and regions of salience network, dorsal attention network and vision. Figure 5(Fig. 5) shows the final group contrasts of seed-to-voxel connectivity maps with ACC seed within different clusters.

ROI-to-ROI analysis of resting state was performed in the default mode regions to investigate passive brain network in the two groups. This analysis showed that there was no significant difference between the two groups in the connectivity of default mode regions (t = -0.268, p = 0.789). Resting state analyses were performed to verify that the brains of the two groups of drivers are normal and healthy, and that there was no significant difference between them without any stimuli or tasks.

After the first level of group analysis (gICA Results), thirty independent components were found based on the defined contrast. To find the components that were significantly different between these two groups, we applied multiple comparison and dual regression on the results of the first level group analysis. The outcome of dual regression analysis showed that there was no component producing significant differences between the two groups. In addition, the independent components didn't pass the dual regression for multiple comparison correction indicating that there was no significant difference in the resting state of brain between the two groups.

Overall, study results indicated higher level of neural activity in the occipital lobe, including lingual gyrus, middle temporal, and lateral occipital cortices, that are related to spatial awareness, perception of distances between environmental elements and motion analysis while performing HPT compared to the other areas of the brain (Figure 2(Fig. 2)).

Traffic information includes awareness about the speed of driver’s own and the surrounding vehicles, the position of driver’s own vehicle, the detection of immediate future hazards (for example, pedestrians, other vehicles and obstacles), and perception of surrounding environment, etc. (Benedikt, 2011[4]). The present study reinforces the role of the neural components of the visual system in safe driving and hazard perception. Integrity of this system has been one of the most important elements for obtaining a license and driving. Our results suggest that neural activity in most of the brain areas related to spatial perception and movement is pivotal for the hazard perception. For example, FEF had significant activity in driving risk perception as it plays an important role in attention, focus, eye movements, and top-down sensory processing in brain's rapid response to external stimuli. Neuronal activation in the FEF leads to enhanced activity of occipital lobe while detecting hazard (Mustari et al., 2009[50]; Schall, 2004[65]). Gaze direction of the eye pupil measured using an eye tracker has been used as parameter for measuring the ability to sense hazard (Caird et al., 2008[14]; Crundall et al., 2004[24]; Land, 2006[43]). However, recently it is suggested that the gaze direction doesn’t mean paying attention and perception of the hazard of the object or subject (Fisher et al., 2016[29]). In addition, different brain areas are activated in perception of driving hazard and are not solely dependent on the activity of the FEF as measured by the eye tracker.

Insular cortex plays an important role in driving hazard management. Brain imaging studies have suggested that its function is related to emotional awareness (Craig and Craig, 2009[20]), interoceptive awareness (Zaki et al., 2012[86]), risk prediction and uncertainty estimation (Bach and Dolan, 2012[3]; Bossaerts, 2010[7]). Insular cortex is also necessary for bottom-up and top down integration and emotional awareness in diverse cognitive tasks (Gu et al., 2013[34]). Thus, it plays a key role in emotional awareness related to a traffic event and contributes to the cognitive control and reaction to the hazardous situations. Activation of insula increases attention necessary for avoiding collisions with approaching objects (Spiers and Maguire, 2007[70]).

Findings of this study showed that parts of the cerebellum play a significant role in the perception of driving hazard. As the primary function of the cerebellum is to coordinate movements of the body and eyes, (Buckner, 2013[10]) this activity may be due to the pupillary movement and visual feedback of the drivers participating in the study while viewing the clips on the scanner (Uchiyama et al., 2003[72]). Activation of cerebellum in cognitive and emotional tasks is reported previously (Adamaszek et al., 2017[1]; Buckner, 2013[10]; Strick et al., 2009[71]), which is consistent with the results of this study.

Overall, our results suggest that situational and emotional awareness are necessary for hazard perception and decision-making in a hazardous situation. In addition, driver’s decision-making based on hazard perception is not only related to approximate information received from images such as motion and cues and prior exposure to the road hazards, which represents a summary of hazardous situations in one's mind. Significant activities of the thalamus, IPS and prefrontal cortex in drivers of this study suggest that the hazard perception skills are also related to the brain regions associated with different attention networks, i.e., alerting, orienting and executive control.

Experienced drivers with better driving hazard perception (both in percentage of correct responses and in reaction time) compared to novices, had significantly greater brain activity in the following regions (Table 3(Tab. 3), Figure 3(Fig. 3)):

Lingual gyrus: It has the V1 (ventral stream) concentrated and is related to awareness about the object and the features of what is being observed (Goel and Dolan, 2001[31]).

As reported previously, experience is an important factor in identifying hazardous situations as it provides better abstract representation (Crundall, 2016[21]; Horswill and McKenna, 2004[39]; Scialfa et al., 2013[66]). Findings of our study have also revealed significant effects of experience on driving hazard perception and it elaborates the role of experience as follows:

In this study, activity of the occipital lobe was significantly higher in experienced compared to inexperienced drivers, indicating that driving experience enhances the skill of having a detailed map of the visual field and the spatial relationships between environmental elements. This skill also requires top-down attention and more pupillary movement, causing higher activity of cerebellum.

In the experienced drivers, the dorsal stream which plays role in spatial representation is strongly connected with FEF and prefrontal cortex regions that are involved in signal detection. This can be a reason for the faster decision-making process during hazard perception among the experienced drivers. Studies using an eye tracker have also shown that focusing on the hazard and pupillary movements of the experienced drivers were significantly greater than inexperienced drivers (Falkmer and Gregersen, 2005[27]).

Our results showed that areas associated with motion perception and behavior prediction were more active in experienced drivers, indicating that one of the important inputs in driving hazard perception is the proper assessment of movement. Distance of an approaching hazard from the vehicle greatly influences the driver's decision making. Accurate estimation of distance and speed is crucial for hazard perception skills. There is a significant relationship between speed and hazard perception (Crundall et al., 2012[23]; Horswill and McKenna, 2004[39]; Renge, 1998[59]). Speed is an influential dimension in responding to road hazards in the HPT (Scialfa et al., 2013[66]). Results of studies on risk perception with what-happen-next approach (Coyne et al., 2007[19]; Lim et al., 2014[44]; Ventsislavova et al., 2016[76]) are consistent with our results suggesting that understanding movement and predicting the near future can be the two very important inputs for deciding how to respond to road hazards. These findings apparently contradict Groeger's (2002[33]) proposed mechanism, suggesting that the experienced drivers in a process without effort and cues recovered from their memory, have better performance in hazard perception.

The AIC plays important role in the emotional awareness and consequence-attention-based decision making (Craig and Craig, 2009[20]; Gu et al., 2013[34]; Zaki et al., 2012[86]), which can lead to faster response speed of experienced drivers in the presence of hazard stimuli in HPT. Perhaps the reason for the significant difference in cerebellar activity between the experienced and novice drivers is the difference in emotional affect resulting from driving images as a driver (Horikawa et al., 2005[36]; Spiers and Maguire, 2007[70]). Age-related alterations in amygdala and cerebellar connectivity from young to middle adulthood on emotion regulation (Xiao et al., 2018[85]) can also explain the results of our study. In addition, behavioral and environmental prediction is reported to be less developed in novice drivers (Vlakveld, 2014[77]). Overall, the findings of our study suggest that experienced drivers with greater brain effort, emotional awareness and cognitive activity have better performance in hazard perception compared to the novice ones.

Novice drivers devote considerable part of their cognitive resources to the vehicle control compared to the visual search by the experienced ones. This results in lower hazard perception and lesser allocation of cognitive resources among the novice drivers that is observed in performing the HPT (Boufous et al., 2011[8]; Brown and Groeger, 1988[9]; Nijboer et al., 2016[52]; Ross et al., 2014[62]).

Results of the present study showed that upper parts of the ACC had more activation in decision-making in response to hazard. These regions are activated in rational cognitive responses (Uchiyama et al., 2003[72]) and ACC activation helps to decrease the effect of distracting factors on executive function (Bush et al., 2000[12]; Spiers and Maguire, 2007[70]). Therefore, responding to the driving hazards is more of a rational cognitive response.

The findings of the present study showed that the connectivity of brain regions related to the executive control attention network (in ACC) differed between the two groups during HPT. As ACC has been found to be highly connected with visual areas of the brain related to perception of space, movement and visual search in experienced drivers (Figure 4(Fig. 4)) (Janzen and Van Turennout, 2004[41]; Malach et al., 1995[46]), it indicates the differences in the mechanism of hazard perception in dividing and focusing attention between the two groups of drivers. The executive control attention network plays an important role in safe driving (Roca et al., 2013[60]; Underwood, 2007[73]) as there is a significant relationship between the occurrence of a driving collision and the executive control attention function (Roca et al., 2013[60]). However, no relationship was previously found between the attention functions and driving performance (Weaver et al., 2009[79]). Our results indicate that the activity of attention-related regions in experienced drivers was significantly higher than novice drivers. In addition, the attention network of the executive control was different between them. This difference in activation of ACC could be due to the negative connectivity with the regions of motion perception and the speed assessment in experienced drivers (Uchiyama et al., 2003[72]). Delay in braking during hazardous situations is attributed to the executive control network (Roca et al., 2013[60]). Based on the results of the present study, the delay in novice drivers can be related to the impact of the negative relationship between ACC and lateral occipital cortices on decision-making (Hahn et al., 2006[35]).

Findings of ICA and DMN analyses of the resting (non-tasked) state of the drivers showed that there were no significant differences in the activities of brain regions and networks between the two groups. In addition, the resting state analysis of the brain showed no difference between the brains of novice and experienced drivers. In addition, their brain activity in the resting state had no superiority over each other. As this study was conducted in a country where hazard perception training is not provided for drivers at any level, and all the participants had no training in this regard. Therefore, the differences found in this study are due to the effect of driving experience.