3. Tests for ADHD diagnostics

Attention tests are much more objective than questionnaires.
On the one hand, however, there is a risk that the results may be distorted. Influences can result from training, hyperfocus of a person with ADHD, test anxiety of a person without ADHD or giftedness. The latter points in particular should therefore always be clarified by means of a differential diagnosis.
On the other hand, their diagnostic value is very limited. Barkley warns in no uncertain terms against the use of neuropsychological tests to diagnose ADHD. Neuropsychological tests - even if they test executive functions - can be successfully completed by more than half of people with ADHD and therefore lead to false non-diagnoses. Rating scales, on the other hand, are much better suited to diagnosing ADHD.¹ Using the NIH Toolbox, Little Man Task, Matrix Reasoning Task and Rey Delayed Recall, one study found a high variance of 31% for ADHD compared to 2.7% for ASD.²

• 1. Test procedure for ADHD diagnostics
  • 1.1. Neuropsychological test procedures
  • 1.2. Individual attention and reaction tests
    • 1.2.1. Stroop Test
    • 1.2.2. Stroop Plus Test
    • 1.2.3. Perceptual Selectivity Test
    • 1.2.4. N-back test
    • 1.2.5. Stop signal task (SST)
    • 1.2.6. Children’s Color Trail Test (1/2)
    • 1.2.7. Continuous Performance Test (CPT)
    • 1.2.8. Digit Span Test
    • 1.2.9. Wisconsin Card Sorting Test
    • 1.2.10. Qb test
  • 1.3. Test matches
  • 1.4. Virtual seminar room (VSR)
• 2. AI test systems: major progress in the lab
• 3. Problems with neuropsychological ADHD tests
  • 3.1. People with ADHD perform as well in tests as people without ADHD with the right motivation
    • 3.1.1. No / fewer ADHD symptoms in tests with high motivation
      • 3.1.1.1. Attention tests
      • 3.1.1.2. Impulsivity (inhibition) can be improved through rewards
      • 3.1.1.3. Reaction times: Reward accelerates reaction times and reduces reaction variance only in ADHD
    • 3.1.2. ADHD symptoms more visible in low-demand tests
  • 3.2. ADHD symptoms and relationship to test subject
  • 3.3. Smoking masks ADHD symptoms
  • 3.4. 116.200 Different ADHD “subtypes
  • 3.5. Evaluations by parents and teachers differ significantly
  • 3.6. Conclusions

1. Test procedure for ADHD diagnostics¶

Source, with more detailed explanations: Schmidt, Petermann.³

1.1. Neuropsychological test procedures¶

• Test battery for attention testing (TAP)
• Attention Load Test - Revision (d2)
• Frankfurt Attention Inventory (FAIR)
• Test battery for career starters - concentration (START-K)
• Frankfurt Adaptive Concentration Test (FAKT-II)
• BLAST (Bron/Lyon Attention Stability Test): Computer-aided test to determine task switching problems⁴
• Cambridge Neuropsychological Test Automated Battery (CANTAB)
  In a study, the CANTAB proved to be unsuitable for the diagnosis of ADHD.⁵
• Conners’ Continuous Performance Test 3rd Edition (CPT3)
• Conners’ Continuous Auditory Test of Attention (CATA)
  CPT3 plus CATA had a higher sensitivity (82.6%), a higher specificity (76%), a higher positive predictive value (88.8%), a higher negative predictive value (65.5%) and an overall higher correct classification rate (80.6%) than CPT3 or CATA alone.⁶

1.2. Individual attention and reaction tests¶

One study found that the Stroop Test, Stroop Plus Test and Perceptual Selectivity Test were good at distinguishing ADHD in adults from non-affected individuals, with the Stroop Test performing best by a narrow margin. However, while the Stroop Test and Stroop Plus Test differed depending on the age of the test subjects, the Perceptual Selectivity Test barely showed any difference across age.⁷ Other studies also showed that the performance of the test subjects in the Stroop Test was age-dependent, with sometimes younger and sometimes older test subjects performing better. In the study mentioned here, both adult persons with ADHD and non-affected people with ADHD performed less well on the Stroop Test as they got older.
The Stroop test is designed to examine selective attention in particular.⁸

1.2.1. Stroop Test¶

Aim of the study: selective attention in the network of the dorsal anterior cingulate cortex (dACC) → striatum → thalamus → dACC.⁸

Color word-color interference test.
4 color words (green, red, blue, yellow) are visually displayed in different colors. The color word and the 4 uncolored answer buttons on the right side of the screen below each other are always in the same place. The color words are displayed for 2 seconds, with a dot to be fixed for 0.75 seconds in between. The word and color match in every fourth display, but not in the others. The respondent is asked to name the visual color of the displayed word as quickly as possible.
If the color word and color are different, this is more difficult. People with ADHD have a higher error rate (in %) and a higher reaction time (in ms).⁷

1.2.2. Stroop Plus Test¶

Aim of the study: selective attention in the network of the dorsal anterior cingulate cortex (dACC) → striatum → thalamus → dACC.⁸

In the Stroop Plus Test, the structure of the Stroop Test is extended by four colored boxes arranged around the word. Together with the word, an arrow appears between the word and the boxes, pointing to one of the colored boxes. The visually displayed color of the word and the color known from the word do not match in 3 out of 4 trials, as in the Stroop test. Only in every ninth trial do the color, word and displayed color box match. The respondent should name the visually displayed color of the word shown as quickly as possible.⁷

1.2.3. Perceptual Selectivity Test¶

Aim of the study: selective attention in the network of the dorsal anterior cingulate cortex (dACC) → striatum → thalamus → dACC.⁸

The Stroop test is presented in the middle of the screen. Four symbols (shape: circles or squares) in different colors (blue and yellow) are displayed around the word, 3 of which show one shape and one the other. In half of the runs, 3 symbols have the same color and one the other, in the other half 2 have one color and 2 the other.
In each of the shapes, a line is displayed that points to 12 o’clock, 1:30 p.m., 3 p.m. or 4:30 p.m. (intended as an hour hand). The test person should only pay attention to the orientation of the line in the symbol that shows the individual shape (which can be horizontal or vertical here) and confirm this as quickly as possible with the assigned button.
This test is designed to measure perceptual selectivity, a subset of selective attention. This term refers to the discriminability of a stimulus, i.e. how effectively the participant can distinguish the target task when confronted with a single stimulus (shape change only) and with two stimuli (shape change and the presence of an irrelevant color).⁷

1.2.4. N-back test¶

Objective: Sustained attention, working memory in the network dlPFC → striatum → thalamus → dlPFC.⁸

N is a variable.
In the visual 0-back test, the respondent is shown a number on a screen and asked to press the key that corresponds to the number. In the 1-back variant, the respondent is asked to press the key of the number that was shown before the one just displayed. In the 2-back variant, the respondent is asked for the penultimate number before the number shown. The higher N is, the greater the load on the working memory.

The N-back test can also be tested by playing an acoustic number announcement. There is also a variant in which different fields light up one after the other on a square field and the respondent has to tap the N field previously activated.

1.2.5. Stop signal task (SST)¶

The stop signal task is used to measure impulsivity or inhibition.

The respondent is asked to record the results of visual symbols that have two alternative outcomes. For example, a left or right button should be pressed as quickly as possible on arrow symbols pointing to the left or right. After a few unevaluated practice runs, the test person should not press a button when an audio signal sounds at the same time, which sounds irregularly for approx. 25% of the runs. The stop signal sounds with a delay after the stimulus. This delay time can vary between 100 ms and 600 ms.⁹

The following are measured

• Number of directional errors
• Proportion of successful stops
• Response time for Go attempts
• Stop signal response time (SSRT)

The stop signals vary in different passes in order to measure the test results of the stop signal response time to 50 % error in order to achieve a measured value that is as comparable as possible. Here is a Video of the stop signal task on Youtube.

The Stop Signal Task is part of the CANTAB. In a study, the CANTAB proved to be unsuitable for the diagnosis of ADHD.⁵

One study converted the stop signal task into an application with mouse control instead of buttons, which showed various improvements in the measurements:¹⁰

• The SSRT shows a weak association with impulsivity, while the measurement of mouse movement shows a strong and significant association with impulsivity
• A machine learning model (weak AI) was able to accurately predict the impulsivity rating of “unknown” participants using mouse movement data from “known” participants
• Mouse movement characteristics such as maximum acceleration and maximum speed are among the most important predictors of impulsivity
• The use of preset stop signal delays leads to a behavior that indicates impulsiveness better than an adaptation of the delay value to previous results (staircase model)

One study found evidence that the performance of stop signal tasks in ADHD reflects impairments in early attentional processes rather than inefficiency in the stopping process.¹¹

1.2.6. Children’s Color Trail Test (1/2)¶

The Children’s Color Trail Test measures attention, divided attention and speed of mental processing and is used, among other things, to diagnose ADHD.¹²¹³

1.2.7. Continuous Performance Test (CPT)¶

A 14-minute continuous performance test (CPT), which requires a high reaction speed, shows variables of improved performance with increasing age:¹⁴

• Response time (RT)
• RT standard error
• Omission error
• Commitment error
• Signal detection parameters (d’ and beta)

The following in particular are gender-specific

• more impulsive mistakes in men
• lower variability in men
• faster RT in men

1.2.8. Digit Span Test¶

The Digit Span Test is part of the Wechsler Adult Intelligence Scale- Revised (WAIS).
The test measures the storage capacity of the working memory for numbers. The test subjects see or hear a sequence of numbers and are asked to reproduce them correctly (forwards or backwards). During the test, the length of the number sequences is increased. The respondent’s number span is the longest number of digits that they can reproduce correctly.
At https://tools.timodenk.com/digit-span-test a Digit Span test can be taken online.

1.2.9. Wisconsin Card Sorting Test¶

The Wisconsin Card Sorting Test is available in both electronic and manual form.¹⁵

1.2.10. Qb test¶

The QbTest combines a continuous performance task (CPT) with a movement tracking system and is intended to support the diagnosis of ADHD. A weak to moderate diagnostic quality was found in children and adolescents.¹⁶

1.3. Test matches¶

• Nesplora Aquarium
  Virtual reality game / test, experimental. It is designed to predict ADHD symptoms in adults and adolescents based on current and retrospective self-reports.¹⁷
• In another approach, artificial intelligence/machine learning was able to correctly diagnose ADHD in players of an online role-playing game (PlayerUnknown’s Battlegrounds) with an 81% probability.¹⁸ A generalized anxiety disorder was detected 84.9% of the time.

1.4. Virtual seminar room (VSR)¶

One study investigated the use of virtual reality (VR) to provide a more realistic and complex, yet standardized, testing environment. In VSR, a virtual continuous performance task (CPT) was combined with simultaneous visual, auditory and audiovisual distractions. At the same time, head movements (actigraphy), gaze behavior (eye tracking), subjective experience, electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) were recorded. Significant differences were found in non-medicated persons with ADHD with regard to¹⁹

• CPT performance
• Head actigraphy
• Distraction gaze behavior
• the subjective experience
  although no group differences were found in theta-beta ratio (EEG) or dorsolateral prefrontal oxyhemoglobin (fNIRS).

2. AI test systems: major progress in the lab¶

Since 2023, AI knowledge systems such as chat GPT have been on everyone’s lips. Machine learning or deep learning are methods for using AI methods

Compared to the varying reliability of mental health diagnoses, the research results in the laboratory already show very impressive values.²⁰
Nevertheless, the results of such test systems are a long way from actually being usable. The detection rates usually only relate to small databases of 20 to several hundred data sets (given in n), consisting exclusively of data from people with ADHD and from (known) healthy subjects. In practice, however, ADHD can also be distinguished from an unknown health status in a mixture of healthy individuals and persons with other mental disorders.
A study that “only” achieved 85% accuracy in 2025, but did not distinguish between ADHD and healthy people, but between ADHD and ASD²¹, should therefore be recognized as a special achievement, even if laboratory studies that distinguish between ADHD and healthy people now exceed 99%.

The temporal and qualitative development of the detection rates gives a good impression of the potential of AI diagnostics. The % figures indicate the accuracy, the year figures indicate the publication date of the descriptive publication.

• 99.58% (2025) by machine learning with the CatBoost algorithm and feature importance scoring with the SHAP algorithm using multidimensional EEG features in children with ADHD compared to healthy controls. ADHD correlated with an increase in the power of theta, alpha and beta rhythms, with increased power ratio between theta and beta (theta/beta ratio, TBR) and with a reorganization of whole brain functional connectivity across all frequency bands (here primarily increased functional connectivity.²³
  99.42% accuracy, 99.03% precision, 99.82% recall, 99.42% F1 score (2025) by EEG signal analysis via deep learning using a ResNet-based model with a doubly extended attention mechanism consisting of autoencoder for feature extraction, Reptile Search Algorithm for feature selection and modified ResNet architecture for model training.²²
• 99.17 % (2024) in a data set of children with ADHD and healthy children using EEG data from the theta band of the frontal and occipital lobe (n = 121)²⁴
• 98.77 % (2023) through a new CNN based on fMRI data sets²⁵
• 98.7% (2024) using video data with facial, postural and hand features of adults with ADHD through machine learning (n = 22)²⁶
  • Accuracy of 98.67 %
  • Precision of 98.01 %
  • Recognition of 98.88 %
• 98.52% (2025) by analyzing EEG, eye-tracking or surveys to detect patterns associated with ADHD symptoms using deep learning using a RhinoFish optimization algorithm to select optimal features (F-score 98.26%, specificity 98.16%)²⁷
• 96.5 % (2024) using the hybrid PUDMO algorithm for the evaluation of EEG data²⁸
• 96 % (2024) by the convolutional neural network CNN model RBP, which was originally developed for the precise classification of brain tumors in medical imaging (n = 3000)²⁹
• 95.3% (2025) by machine learning using the Isolation Forest algorithm and subsequently a recurrent neural network (RNN) model, using the HYPERAKTIV dataset, which contains high-resolution temporal data on the motor activity of people diagnosed with ADHD.³⁰
• 94.5 % (2024) with the LRP algorithm (Layer-wise Relevance Propagation) based on EEG data³¹
• 93.9 % (2024) Frequency-Integrated Visual-Language Network (FIVLNet), a deep learning framework for analyzing MRI scans³²
• 92 % (2025) Machine learning model of analyzing written natural language reports from patients who had undergone intelligence testing. Enabled the reconstruction of human-readable text from the selected N-gram features.³³
• 90.81 % (2025) Machine learning from EEG data of n = 2434 19-channel EEG data recordings from 5 frequency bands of n = 168 subjects³⁴
• 90.02 % (2021) accuracy, AUC 95.7 %, sensitivity 89.8 % and specificity 90.55 % achieved an analysis of ADHD gene candidates using deep learning (“AI”) by combining genes with insignificant P-values.³⁵
• 89.89% (2025) by near-infrared spectroscopy (fNIRS) during verbal tasks in conjunction with machine learning (ML) to discriminate between adults with ADHD (n = 120) and healthy controls (n = 75)³⁶
• 89 % (2025) Random forest model based on EEG data³⁷
• 88.6 % (2024) Random forest model for resting-state theta-band EEG data³⁸
• 88 % (2020) using a convolutional neural network (CNN) to analyze EEG data, in particular event-related potentials collected during the flanker task³⁹
• 82 % (2023) accuracy for a special AI system for the diagnosis of ADHD. The sensitivity was 64 % and the specificity 99 %. The accuracy was higher than for all other disorders tested with this system⁴⁰
• 79 % (2023) through a combination of 4 machine learning algorithms for analyzing register data.⁴¹
• 75.6% (2021) using a combination of Convolutional Denoising Autoencoders (CDAE) classification method and Random Forest (RF) algorithms, with a sensitivity of 76.9% and a specificity of 73.1%⁴²
• 74.9% (2020) by a multimodal 3D-CNN using gray matter features and fALFF from fMRI⁴³
• 69.15 % (2017) using a deep learning model to classify MRI scans⁴⁴

Further AI systems for diagnosing ADHD are currently under development.⁴⁵

3. Problems with neuropsychological ADHD tests¶

When testing people with ADHD, it must always be taken into account that the symptoms to be measured are not always and consistently present in people with ADHD. Rewards change the symptom pattern in ADHD.⁴⁶ In particular, a high motivation of the person with ADHD to participate in the test (intrinsic or through high extrinsic incentives) can distort the test results. This can go so far that the test is negative even though ADHD is present.

3.1. People with ADHD perform as well in tests as people without ADHD with the right motivation¶

Motivation is altered in people with ADHD. Extrinsic incentives must be higher than for those without the disorder in order to trigger the same motivation.

Against this background, computer games that aim to measure ADHD could face the problem that the more fun the computer game is, the less meaningful the test results become. One study did not find significantly higher levels of flow and intrinsic motivation in a stop-signal game than in a non-gameified stop-signal test.⁴⁷

3.1.1. No / fewer ADHD symptoms in tests with high motivation¶

3.1.1.1. Attention tests¶

In ADHD, it is not the technical ability to pay attention or direct attention that is impaired, but the way attention is directed is subject to different criteria than in those not affected. Attention control in ADHD is subject to a different control system, which is much more dependent on the satisfaction of one’s own needs than in non-affected people. However, this is not selfishness, but an altered control of motivation over which the person with ADHD has no influence.

If personal needs are addressed through intrinsic interest or rewards (which must be higher on average than for people not affected), the attention performance of people with ADHD improves more than that of those not affected.⁴⁸ Ryffel-Rawak describes a typical example.⁴⁹
A meta-study confirms that high rewards greatly improve test scores in ADHD.⁵⁰

In ADHD, deactivation of the default mode network (DMN), as occurs with attention to the environment, is significantly reduced. Personal motivation induced by high rewards equalized DMN deactivation in ADHD, so that it corresponded to that of non-affected individuals. MPH had the same effect.⁵¹

3.1.1.2. Impulsivity (inhibition) can be improved through rewards¶

Impulsivity in ADHD (inhibition ability) can also be influenced by rewards,⁵³ up to results equivalent to controls, while the same rewards improved inhibition less in children with brain injuries.⁵⁴ Another study also found identical inhibition ability of persons with ADHD and non-affected people with high rewards.⁵⁵ Another study found no improvement.⁵⁶ It would be necessary to check whether the rewards here were possibly too low to arouse personal motivation.

3.1.1.3. Reaction times: Reward accelerates reaction times and reduces reaction variance only in ADHD¶

The test results on reaction times in ADHD are inconsistent. Many tests show a reduced reaction time in people with ADHD, while individual tests show an increased reaction time. ⇒ Reduced reaction time with ADHD?

On average, the reaction time of people with ADHD varies more than that of people without ADHD.
⇒ Reaction time variance increased in ADHD

In one study, when rewarded, only the persons with ADHD and their unaffected siblings showed accelerated mean reaction times and lower response variability, but not the controls, while accuracy improved in all 3 groups.⁵⁷

3.1.2. ADHD symptoms more visible in low-demand tests¶

Activating, exciting tasks can cause test results of people with ADHD to correspond to those of controls.⁵⁸ For example, test results of people with ADHD show inconspicuous results when a high test event rate (fast test) is presented, while a low event rate (slow test) shows significant differences.⁵⁹⁶⁰

One study hypothesizes that a slower event rate on tests in people with ADHD causes reduced intrinsic motivation due to their specific motivational characteristics.⁶¹

The optimal arousal range, in which no under- or overstimulation occurs, is much narrower with ADHD than with non-affected people.
Even more important than the optimal arousal range, however, is the fact that persons with ADHD can direct their attention just as well as people without ADHD if they are personally motivated (example: hyperfocus). This is understandable in light of the stress benefits of altered attention control in ADHD.
⇒ Stress benefits of distractibility, task switching problems and attention problems

In addition to under- and overload, this limited capacity can also exist in terms of time, in the sense that the ability to compensate for under- or overload is limited. One teacher described her perception of a child with ADHD-I to us by saying that she could see how the child “ran full” in the first lesson and then developed the typical ADHD-I symptoms.

3.2. ADHD symptoms and relationship to test subject¶

One study addresses how the relationship with the test administrator influences the test results of people with ADHD.⁶³

3.3. Smoking masks ADHD symptoms¶

Smoking as self-medication increases dopamine levels - albeit only in the short term. It also reduces stress symptoms and irritability. People with ADHD smoke about twice as often as people without ADHD.
Smoking can therefore make the diagnosis of ADHD more difficult.⁶⁴

A study of emotional dysregulation in smokers with ADHD found no differences between those who smoked as usual and those who abstained from smoking for 24 hours.⁶⁵
It is well known that withdrawal from alcohol and other drugs requires a period of 2 to 4 weeks so that the receptors downregulated by the high supply of neurotransmitters (dopamine, etc.) can upregulate again. This is responsible for the withdrawal symptoms, among other things. This also applies to nicotine. It is possible that only tests that require nicotine withdrawal of more than 2 to 4 weeks will be able to show a difference.

3.4. 116.200 Different ADHD “subtypes¶

One study calculated that there are 116,200 different ADHD subtypes - if each combination of symptoms is considered a separate subtype.⁶⁶

Against this background, the currently (still?) common test procedures, which (using DSM 5 as an example) are based on 9 characteristics for hyperactivity/impulsivity and 9 characteristics for inattention, of which at least 6 characteristics must be fulfilled in at least one category, should be questioned. People with ADHD with 5 features in both categories would be qualified by DSM 5 as “no ADHD”. Not all doctors realize that the DSM and ICD are merely guidelines and not diagnostic standards - a misunderstanding that Francis Allen, the head of the DSM IV Commission, warned against.
On this basis, we consider Barkley’s and our approach of a diagnosis based on a query of a large number of symptoms to be well worth considering. ⇒ The big ADxS.ORG - ADHD online test

3.5. Evaluations by parents and teachers differ significantly¶

One study found that teachers’ and parents’ evaluations of the effects of medication for ADHD did not correlate. These strong differences demonstrate that different sources should be used when evaluating ADHD.⁶⁷

3.6. Conclusions¶

For reliable results, tests for ADHD would have to be carried out correctly in different test environments that take into account the degree of underchallenge, appropriate arousal and overchallenge. Furthermore, smoking behavior should be included in the weighting of the test results, which has not been done sufficiently to date. Testing on the basis of all possible ADHD symptoms seems to us to be an option worth considering.