**5. Discussion**

#### **5.1. Part 1: self-reporting types, priorities, and characteristics**

The multiphase structure of our study was instrumental in translating our interviews, literature review, and expert panel findings into effective online survey questions. The key findings included the types, priorities, and characteristics of self-reported data that clinicians need from patients as shown in **Figure 2**.

The remainder of this section highlights notable patient self-reporting challenges as well as subsequent feedback after sharing these findings with clinicians.

#### *5.1.1. Self-reporting availability*

Many symptoms and triggers were reported as "useful" but "unavailable" as shown in orange in **Figure 2**. "Academic decline" was said to be unavailable (five out of six respondents). These findings highlight the need for patient data that may already be collected but unavailable to clinicians. Improved interoperability between electronic health records (EHRs) and electronic grading systems could alert clinicians to changes in patient grades during appointments.

#### *5.1.2. Self-reporting difficulty*

**Modality Mean SD Min Max Right-tail hypothesis test**

Daytime reporting [9] 0.810 100.00 68.00 91 Patient Nighttime reporting [9] 0.253 100.00 14.50 91 Patient

**Systems F-score Precision Recall PWS Modality** Van de Vel [82] 0.721 0.578 0.957 7 Inertial Conradsen [63] 0.682 0.750 0.625 2 EMG Jallon [78] 0.639 0.717 0.577 2 Inertial Narechania [64] 0.580 0.430 0.890 13 Pressure Van de Vel, VARIA [65] 0.560 0.560 0.560 1 Multimodal Poh [41] 0.508 0.349 0.938 7 Multimodal Nijsen [84] 0.492 0.350 0.830 18 Inertial Nijsen [83] 0.487 0.350 0.800 36 Inertial Bruijne Screams [61] 0.459 0.300 0.980 17 Audio Van de Vel, Epi-care Free [65] 0.410 0.410 0.410 1 Inertial Van de Vel, Epi-care [65] 0.400 0.400 0.400 1 Inertial van Elmpt [62] 0.391 0.900 0.250 3 ECG

Carlson MP5 [55] 0.385 0.278 0.625 4 Multimodal Pisani [47] 0.201 0.117 0.714 12 Video Lockman [42] 0.133 0.072 0.875 6 Inertial Fulton MP5 [54] 0.083 1.000 0.043 15 Multimodal Fulton ST-2 [54] 0.043 1.000 0.022 15 Inertial Bruijne Lip smacking [61] 0.039 0.020 0.980 17 Audio

Inertial 0.598 0.282 0.043 0.99 0.008 1 Video 0.786 0.292 0.201 0.964 0.426 0.997 Pressure 0.68 0.142 0.58 0.78 0.209 0.927 EMG 0.682 0 0.682 0.682 – – Multimodal 0.384 0.214 0.083 0.56 0.014 0.846 Audio 0.249 0.297 0.039 0.459 0.114 0.494 ECG 0.391 0 0.391 0.391 – – All systems 0.585 0.288 0.039 0.99 0 1

**Table 3.** System F-score and p-value statistics by modality.

**Table 2.** System and patient self-reporting performance comparison.

**Self-reporting**

82 Seizures

**p-Value day p-Value night**

There were several symptoms and triggers that were reported as "difficult" for patients to report as shown in yellow in **Figure 2**. Notably, "Seizure onset at night" and "excessive sleep movements" were said to be "difficult" to report among most epilepsy specialists (five out of six respondents). These findings highlight the inherent difficulty of patient data collection while sleeping or unconscious. Introducing automated wrist-worn devices such as the Empatica E4 [68] and ActiGraph Link [69] could stand to increase patient self-reporting performance by detecting events such as seizure and unusual sleep movements, while also reducing patient and caregiver data collection burden.

#### *5.1.3. Self-reporting reliability*

Next, many symptoms and triggers were reported as being useful but unreliable when selfreported as shown in gray in **Figure 2**. All epilepsy specialists (six out of six respondents) agreed that patient and caregiver reports of patient "memory impairment" were "unreliable". These findings suggest a need for more reliable and simple patient data collection tools. For example, introducing automated data collection tools could help to increase reliability for clinicians by making data collection more accurate and consistent. The forward auditory Digit Span task, WISC-R subtest [70] could be administered as a smartphone unlock screen for periodically assessing short-term memory.

**5.2. Part 2: seizure reporting performance and capabilities**

perceived accuracy of patient reports.

accurate seizure counts over time.

ties for informing clinical decisions.

evaluating the validity of patient reports.

*5.2.1.1. Informing initial AED selection*

*5.2.1. Self-reporting needs*

Interviews and our literature review and card sorting exercise with clinicians helped us characterize the types of information that neurologists deem to be the most important during typical stages of epilepsy treatment, how likely they are to have access to this information, and the

Self-Reporting Technologies for Supporting Epilepsy Treatment

http://dx.doi.org/10.5772/intechopen.70283

85

Most neurologists reported having access to EEG reports and verbal descriptions of seizures during treatment. However despite this information, neurologist also expressed a need for more and more detailed characterization of patient movement during seizures and more

These needs were then further reflected in our literature review as we explored current methods for characterizing motion during seizures and compared existing patient seizure counting performance to current seizure detection systems. Moreover, our online survey results highlighted two important self-reporting challenges. First, there are limited recording and annotation tools available for characterizing patient motion during seizures. Second, seizure detection systems tend to have false positives and therefore over report seizures. Introducing video capture systems that are triggered by wearable seizure detection sensors may prove beneficial in both cases. More accurate seizure data could, therefore, present new opportuni-

Neurologists reported mixed reliance on patient and caregiver reports when making decisions during treatment. In our questionnaire, 70% of neurologists rated patient and caregiver self-reporting as playing a significant role when determining the best course of AED treatment (4 or greater on a scale of 5), however there was considerable in terms of how frequently these initial self-reports included patient movement characteristics during seizures (SD = 1.10) and/or described the evolution of the seizure over time (SD = 0.78). This finding suggests that, while neurologists perceive self-reporting as important, they also emphasize the need for

Neurologists from our survey indicated that support for characterizing patient seizure type could be beneficial for selecting the most suitable initial AED based on the patient's seizure symptoms. The survey respondents ranked seizure type and movement characterization as the most important information during the initial diagnosis and AED selection phase. Most respondents had access to EEG reports (7/10) and MRI reports (5/10) and verbal accounts of seizures (80%). Less than one-third of neurologists had access to hospital records, imaging records, blood work, seizure diaries, and video of patient seizure events. Most notably, while all neurologists (10/10) expressed a desire for supplemental video only 3/10 respondents had

regular access to such video for informing diagnosis and treatment.

#### *5.1.4. Self-reporting desired frequency*

Finally, the majority response for desired patient self-reporting frequency is shown in each column of **Figure 2**. Epilepsy clinicians desired daily reports for over half of all "top 20" indicators (11 out of 20 items). New onset "viral infections" and "status seizures" require immediate medical attention for managing seizure control medications.

Finally, "episodic" patient self-reporting was most frequently desired by psychologists. Many patient behavior changes are highly context driven such as "loss of interest in activities". These findings have implications for displaying patient health dashboards for clinicians within these respective specialties as episodic changes may be more difficult to anticipate than daily, weekly, and monthly data collection on a pre-defined schedule.

#### *5.1.5. Self-reporting challenges*

Mental health indicators and patient seizure reports may not always be available to clinicians during pediatric epilepsy treatment.

Most notably, suicide attempts were reported as useful, unavailable, and difficult for patients to collect at monthly intervals by all but one clinician. Medication side effects can often trigger depression and there is a high prevalence of depression among patients with epilepsy. If symptoms are known, then clinicians can consider prescribing a seizure control medication that may be less effective but help to stabilize mood. The main challenge for clinicians is that this type of data is often not available or difficult to collect depending on the patient's age and caregiver situation. For example, a primary caregiver may be knowledgeable of the patient's mental health but a patient may be accompanied by an uninformed family member.

The onset of nighttime seizure reports was also reported as useful, unavailable, and difficult for patients to collect at daily intervals by most clinicians. All respondents indicated that patient reports were unreliable due to age and cognition. This reliability is important for treating and thereby helping to reduce the risk of a condition called sudden unexpected death in epilepsy (SUDEP).

These results suggest the need for more reliable and easy to use mental health and seizure reporting tools. For example, automated weekly or monthly validated mental health surveys such as the Personal Health Questionnaire PHQ-9 [71] could be emailed or assessed in clinic on a tablet to increase reliability when it comes to screening for suicide attempts and depression. Finally, clinicians could suggest that patients wear seizure detection wristbands at night for detecting and reporting convulsive type seizures [41].

#### **5.2. Part 2: seizure reporting performance and capabilities**

Interviews and our literature review and card sorting exercise with clinicians helped us characterize the types of information that neurologists deem to be the most important during typical stages of epilepsy treatment, how likely they are to have access to this information, and the perceived accuracy of patient reports.

Most neurologists reported having access to EEG reports and verbal descriptions of seizures during treatment. However despite this information, neurologist also expressed a need for more and more detailed characterization of patient movement during seizures and more accurate seizure counts over time.

These needs were then further reflected in our literature review as we explored current methods for characterizing motion during seizures and compared existing patient seizure counting performance to current seizure detection systems. Moreover, our online survey results highlighted two important self-reporting challenges. First, there are limited recording and annotation tools available for characterizing patient motion during seizures. Second, seizure detection systems tend to have false positives and therefore over report seizures. Introducing video capture systems that are triggered by wearable seizure detection sensors may prove beneficial in both cases. More accurate seizure data could, therefore, present new opportunities for informing clinical decisions.

#### *5.2.1. Self-reporting needs*

These findings suggest a need for more reliable and simple patient data collection tools. For example, introducing automated data collection tools could help to increase reliability for clinicians by making data collection more accurate and consistent. The forward auditory Digit Span task, WISC-R subtest [70] could be administered as a smartphone unlock screen for peri-

Finally, the majority response for desired patient self-reporting frequency is shown in each column of **Figure 2**. Epilepsy clinicians desired daily reports for over half of all "top 20" indicators (11 out of 20 items). New onset "viral infections" and "status seizures" require immedi-

Finally, "episodic" patient self-reporting was most frequently desired by psychologists. Many patient behavior changes are highly context driven such as "loss of interest in activities". These findings have implications for displaying patient health dashboards for clinicians within these respective specialties as episodic changes may be more difficult to anticipate

Mental health indicators and patient seizure reports may not always be available to clinicians

Most notably, suicide attempts were reported as useful, unavailable, and difficult for patients to collect at monthly intervals by all but one clinician. Medication side effects can often trigger depression and there is a high prevalence of depression among patients with epilepsy. If symptoms are known, then clinicians can consider prescribing a seizure control medication that may be less effective but help to stabilize mood. The main challenge for clinicians is that this type of data is often not available or difficult to collect depending on the patient's age and caregiver situation. For example, a primary caregiver may be knowledgeable of the patient's

mental health but a patient may be accompanied by an uninformed family member.

for detecting and reporting convulsive type seizures [41].

The onset of nighttime seizure reports was also reported as useful, unavailable, and difficult for patients to collect at daily intervals by most clinicians. All respondents indicated that patient reports were unreliable due to age and cognition. This reliability is important for treating and thereby helping to reduce the risk of a condition called sudden unexpected death in

These results suggest the need for more reliable and easy to use mental health and seizure reporting tools. For example, automated weekly or monthly validated mental health surveys such as the Personal Health Questionnaire PHQ-9 [71] could be emailed or assessed in clinic on a tablet to increase reliability when it comes to screening for suicide attempts and depression. Finally, clinicians could suggest that patients wear seizure detection wristbands at night

ate medical attention for managing seizure control medications.

than daily, weekly, and monthly data collection on a pre-defined schedule.

odically assessing short-term memory.

*5.1.4. Self-reporting desired frequency*

84 Seizures

*5.1.5. Self-reporting challenges*

epilepsy (SUDEP).

during pediatric epilepsy treatment.

Neurologists reported mixed reliance on patient and caregiver reports when making decisions during treatment. In our questionnaire, 70% of neurologists rated patient and caregiver self-reporting as playing a significant role when determining the best course of AED treatment (4 or greater on a scale of 5), however there was considerable in terms of how frequently these initial self-reports included patient movement characteristics during seizures (SD = 1.10) and/or described the evolution of the seizure over time (SD = 0.78). This finding suggests that, while neurologists perceive self-reporting as important, they also emphasize the need for evaluating the validity of patient reports.

#### *5.2.1.1. Informing initial AED selection*

Neurologists from our survey indicated that support for characterizing patient seizure type could be beneficial for selecting the most suitable initial AED based on the patient's seizure symptoms. The survey respondents ranked seizure type and movement characterization as the most important information during the initial diagnosis and AED selection phase. Most respondents had access to EEG reports (7/10) and MRI reports (5/10) and verbal accounts of seizures (80%). Less than one-third of neurologists had access to hospital records, imaging records, blood work, seizure diaries, and video of patient seizure events. Most notably, while all neurologists (10/10) expressed a desire for supplemental video only 3/10 respondents had regular access to such video for informing diagnosis and treatment.

These findings stress a need for capturing additional patient information prior to diagnosis and have implications for patient and caregiver data collection efforts. MRI and EEG may not be available for first-time general practitioner referrals. Initial outpatient EEG sessions tend to be short, ~20 min. Moreover, even with routine activation procedures such as patient hyperventilation, photic stimulation, and sleep withdrawal, many patients may not show symptoms during a single visit and require further observation. It may, therefore, be helpful for patients to collect additional seizure observations such as video recordings prior to initial appointments.

precision, but fewer numbers of patients [45, 46, 73]. In turn, more work is needed for reducing false positives among all classes that we surveyed. **Table 3** shows that many of the best performing systems utilize video with an average F-score of 0.79 (SD = 0.29) while audiobased systems performed the worse with an average F-score performance of 0.25 (SD = 0.30) with precision as low as 2% for detecting audible lip smacking [61]. High p-values above 0.05 in **Table 3** suggest that mean F-scores for each type of system share a greater than chance probability of performing better than self-reporting at night while low p-values suggest that

Self-Reporting Technologies for Supporting Epilepsy Treatment

http://dx.doi.org/10.5772/intechopen.70283

87

Most systems performed better than patient reporting during the night but notably worse than patients during the day. In our review, all but two systems achieved higher F-score performance at night while only four inertial systems performed better during the day. The average F-score for all systems was 0.59 (SD = 0.29); this reflects a notable improvement over patient self-reporting at night (F-score 0.25) yet remains significantly worse than self-reporting during the day (F-score 0.81). Inertial systems are shown to perform well across both day and nighttime studies [41, 43], however, as noted more work must be done for reducing false alarms during daily activities [40]. High-performance variability was observed between the same types of systems. These dis-

**1. Day versus night:** Many systems were only evaluated at night [55, 65, 73], or strictly during the day, [36] while others were evaluated during the night and day [41–43]. Nighttime studies tended to perform better than daytime studies with an average F-score of 0.62 as compared to 0.56 during studies that included daytime monitoring. This makes direct comparison difficult because daytime seizure detectors must also distinguish non-seizure events such as exercise and teeth brushing which were less prevalent at night. For example, Van De Vel et al.'s [65] evaluation of the Emfit pressure mat highlighted false positives when sitting up in bed, Pisani et al. [47]'s video analysis confused random infant movements, Lockman et al.'s inertial wristband [42] reported false positives during rhythmic activities such as brushing teeth and pen tapping while Poh et al. [41] reported similar false

**2. Alerting versus reporting:** Many systems are primarily designed for alerting caregivers rather than accurately reporting seizure counts. Existing commercial systems are designed for alerting caregivers to ongoing seizures [43, 53, 67]. The caregiver is often burdened with adjusting system-specific threshold settings for minimizing false positives [51, 74–75].

**3. Patient age:** We observed considerable variation between the age and number of patients enrolled in studies. For example, Cuppens et al. [73] and Lockman et al. [42] each developed similar inertial-based systems, however, Cuppens et al. [73] studied patients aged 5–16 while Lockman et al. [35] studied ages 3–85. It may be reasonable to expect that differences in muscle development and limb length between these age groups could have

**4. Patient count:** The number of patients with seizures also varied between studies with a single patient having seizures at night. For example, seven studies had less than four

This, in turn, may result in missing facial tics and other less apparent symptoms.

resulted in slightly different movement characteristics during seizures.

systems will perform worse than patients during the day on average.

crepancies can largely be explained by the following four contributing factors:

positives during dice rolling and video game activities.

#### *5.2.1.2. Informing AED adjustment*

Neurologists ranked seizure frequency as the most important patient self-reported information available to them (100% rated 5 out of a scale of 5) for making AED adjustments. Most neurologists (8/10) estimated that patients failed to report between 40 and 60% of seizures overall (given 5 uniform ranges between 0 and 100%). This estimate agreed with Hoppe et al.'s findings that 55% of patients failed to document 55% of seizures overall [9]. Most neurologists also agreed that an ictal description of a seizure is the most difficult for a patient to report, and 66% of the surveyed neurologists said that patients or caregivers report less than 60% of their seizures. It may, therefore, be helpful to introduce seizure detection devices that address specific patient challenges such as nighttime reporting.

#### *5.2.2. Seizure reporting shortcomings*

Major shortcomings of current seizure classification and detection technologies include (1) limited capture and playback solutions for characterizing seizure type and (2) inaccurate seizure detection for counting seizures and limited support for identifying seizure types that do not exhibit limb movement.

#### *5.2.2.1. Limited tools for AED selection*

The prospect of developing motion characterization tools for informing initial AED selection remains largely unexplored. Efforts have been limited to active and passive motion tracking as additional feedback for EEG technicians [38, 51]. To date, existing research and commercial systems have not focused on the problem of motor characterization for initial partial versus generalized seizure characterization. There is, therefore, a need to utilize additional video and motion tracking technologies for informing AED selection.

#### *5.2.2.2. Inaccurate seizure counts for AED adjustment*

Neurologists from our survey agreed that accurate seizure counts are the most important feedback. In our review, inertial seizure detection systems [43, 45, 46, 72] achieved higher performance than embedded mattress devices [55, 65] and multimodal devices [41]. Inertial devices also tend to support daytime use, while mattress and video systems are often limited nighttime use within bedrooms [47, 65].

High false positive rates remain a problem. More accurate seizure counts could better inform AED treatment. We contend that false positives remain a problem despite studies with higher precision, but fewer numbers of patients [45, 46, 73]. In turn, more work is needed for reducing false positives among all classes that we surveyed. **Table 3** shows that many of the best performing systems utilize video with an average F-score of 0.79 (SD = 0.29) while audiobased systems performed the worse with an average F-score performance of 0.25 (SD = 0.30) with precision as low as 2% for detecting audible lip smacking [61]. High p-values above 0.05 in **Table 3** suggest that mean F-scores for each type of system share a greater than chance probability of performing better than self-reporting at night while low p-values suggest that systems will perform worse than patients during the day on average.

These findings stress a need for capturing additional patient information prior to diagnosis and have implications for patient and caregiver data collection efforts. MRI and EEG may not be available for first-time general practitioner referrals. Initial outpatient EEG sessions tend to be short, ~20 min. Moreover, even with routine activation procedures such as patient hyperventilation, photic stimulation, and sleep withdrawal, many patients may not show symptoms during a single visit and require further observation. It may, therefore, be helpful for patients to collect additional seizure observations such as video recordings prior to initial appointments.

Neurologists ranked seizure frequency as the most important patient self-reported information available to them (100% rated 5 out of a scale of 5) for making AED adjustments. Most neurologists (8/10) estimated that patients failed to report between 40 and 60% of seizures overall (given 5 uniform ranges between 0 and 100%). This estimate agreed with Hoppe et al.'s findings that 55% of patients failed to document 55% of seizures overall [9]. Most neurologists also agreed that an ictal description of a seizure is the most difficult for a patient to report, and 66% of the surveyed neurologists said that patients or caregivers report less than 60% of their seizures. It may, therefore, be helpful to introduce seizure detection devices that

Major shortcomings of current seizure classification and detection technologies include (1) limited capture and playback solutions for characterizing seizure type and (2) inaccurate seizure detection for counting seizures and limited support for identifying seizure types that do

The prospect of developing motion characterization tools for informing initial AED selection remains largely unexplored. Efforts have been limited to active and passive motion tracking as additional feedback for EEG technicians [38, 51]. To date, existing research and commercial systems have not focused on the problem of motor characterization for initial partial versus generalized seizure characterization. There is, therefore, a need to utilize additional video and

Neurologists from our survey agreed that accurate seizure counts are the most important feedback. In our review, inertial seizure detection systems [43, 45, 46, 72] achieved higher performance than embedded mattress devices [55, 65] and multimodal devices [41]. Inertial devices also tend to support daytime use, while mattress and video systems are often limited

High false positive rates remain a problem. More accurate seizure counts could better inform AED treatment. We contend that false positives remain a problem despite studies with higher

address specific patient challenges such as nighttime reporting.

motion tracking technologies for informing AED selection.

*5.2.2.2. Inaccurate seizure counts for AED adjustment*

nighttime use within bedrooms [47, 65].

*5.2.1.2. Informing AED adjustment*

86 Seizures

*5.2.2. Seizure reporting shortcomings*

*5.2.2.1. Limited tools for AED selection*

not exhibit limb movement.

Most systems performed better than patient reporting during the night but notably worse than patients during the day. In our review, all but two systems achieved higher F-score performance at night while only four inertial systems performed better during the day. The average F-score for all systems was 0.59 (SD = 0.29); this reflects a notable improvement over patient self-reporting at night (F-score 0.25) yet remains significantly worse than self-reporting during the day (F-score 0.81). Inertial systems are shown to perform well across both day and nighttime studies [41, 43], however, as noted more work must be done for reducing false alarms during daily activities [40].

High-performance variability was observed between the same types of systems. These discrepancies can largely be explained by the following four contributing factors:


participants [45, 62, 63, 65, 76–78]. Van De Vel et al. [65] and Narechanie et al. [64] each evaluated pressure sensing mattress inserts, however, Van De Vel et al. [65] included only 1 patient with an F-score of 0.78 while Narechanie et al. [64] included 51 patients with a perfect F-score of 1.0 for reporting seizure counts at night.

and sleep indicators during epilepsy treatment. In our technology review, we surveyed seven types of seizure detection sensing modalities and identified a strong need for more accurate and reliable seizure reporting and motion characterization during diagnosis and

Self-Reporting Technologies for Supporting Epilepsy Treatment

http://dx.doi.org/10.5772/intechopen.70283

89

**1.** Identifying the specific types, priorities, and characteristics of data that clinicians need

**2.** Establishing the extent that current health tracking devices are suitable for addressing

The findings from our research highlighted important patient self-reporting needs among a diverse set of clinicians for epilepsy diagnosis and treatment and in turn, may provide clinicians and technology developers with a useful reference for aligning development efforts with clinical information needs within epilepsy treatment. High false positives remain a problem for seizure detection devices; however low-cost hardware may be able to mitigate these issues. For example, inertial sensors [42] and the bedroom instrumented camera [49] could be sent home with patients prior to treatment; such a combination of tools could be vital infor-

This chapter is taken in part from our paper on, "Seizure reporting technologies for epilepsy

, Cherise Frazier2

and Elizabeth D. Mynatt<sup>1</sup>

treatment: A review of clinical information needs and supporting technologies" [24].

, Cam Escoffery<sup>3</sup>

1 Georgia Institute of Technology, School of Interactive Computing, Atlanta, GA, USA

[1] Kooiman TJM, Dontje ML, Sprenger SR, Krijnen WP, van der Schans CP, de Groot M. Reliability and validity of ten consumer activity trackers. BMC Sports Science,

3 Emory University, Rollins School of Public Health, Atlanta, GA, USA

The key challenges faced by technology developers and providers are:

treatment.

from patients.

**Acknowledgements**

**Author details**

Jonathan Bidwell1

**References**

these needs is similarly unknown.

mation aiding neurologist in how best to treat the patient.

\*, Eliana Kovich2

\*Address all correspondence to: bidwej@gatech.edu

Medicine and Rehabilitation Oct 2015;**7**:24

2 Children's Healthcare of Atlanta, GA, USA

#### *5.2.2.3. Limited diversity of seizure types*

Most patients have focal types seizures (70%>50%) [80, 81]; however, only some but not all focal seizures involve limb movement. This presents a challenge as most systems to date are limited to measuring seizures based on limb movements. More reliable metrics or a combination of metrics should be studied for capturing non-motor seizure symptoms.

To date, there has been limited work on detecting seizures using non-inertial and video sensors. Bruijne et al. analyzed [61] audio for detecting "lip smacking" and "screams" however; the performance was among the poorest of all the systems that we evaluated (F-score = 0.04). To the best of our knowledge, there are no non-EEG devices for detecting symptoms (e.g. subtle face or hand movement during partial seizures or behavioral arrest).

Inertial seizure detection wristbands [43] and nighttime video recording could provide a promising short-term solution for increasing the accuracy of patient reporting. Most patients are seen by a general practitioner and are later referred to see a neurologist [79]. This gap presents an opportunity to equip patients with data collection systems for detecting and recording patient seizures in the home prior to an initial neurology visit. For example, the open source OpenSeizureDetector [52] inertial wristband could be used in conjunction with an already available and bedroom instrumented camera such as the SAMi [49] or OpenSeizureDetector [50] detect seizures and trigger video recording. In turn, neurologists could review video for characterizing the seizure prior to treatment. Movements during seizures could be captured and reviewed.

Finally, seizure reporting video annotation tools could enable patients, caregivers, and neurologists to label the start and stop of seizure events could improve seizure detection performance over time and address the problem of having to manually adjust thresholds as in commercial products [42, 43] as the system will be trained for a particular individual.
