Major Depressive Disorder (MDD) is a serious mental disorder that afects millions of adults, occasionally leading to life-threatening results. Current diagnostic tools for MDD mostly consist of questionnaires and/or long, specialized therapy sessions. In this work we present a serious game called "The Delivery", developed for diagnosing MDD in players. The video game has the players immerse into a realistic scenario, the development of which depends on their actions, that is, through conversations with in-game characters, completion of quests, and interactions with the environment. All in-game features and mechanics are designed to correspond to specific diagnostic criteria for MDD. We recorded gameplay data from labeled players (both MDD and non-MDD cases) in order to train Machine Learning models that can accurately distinguish gameplay behaviors MDD-positive and MDD-negative players.
• Computing methodologies → Machine learning; • Applied computing → Life and medical sciences.
Major Depressive Disorder (MDD) is the most common type of depression in adults. The characteristics of MDD can be really hard to identify and manage, especially due to the fact that they are triggered by psychological factors. Since identifying the exact pathological and psychological features in a patient is dificult, alternative methodologies for this disorder’s diagnosis should be explored.
This poses a challenge that many scientists attempt to tackle in
various ways. Serious games (i.e. games with a specific purpose
other than entertainment) have also made a step towards the
improvement of mental health issues [
Other serious games exist for the treatment or prevention of
depression [
Moreover, our work is in line with the directions proposed by
[
In [
This section covers the basic steps taken in regard to extracting the resulting conclusions. These steps include data processing operations, as well as algorithm selection, implementation and tuning carried out for the experiments. 2.1
Initially we developed a serious, story-based game called "The Delivery"1, for the diagnosis of Major Depressive Disorder (MDD) using Artificial Intelligence and Machine Learning methods. "The Delivery" follows a relatively realistic scenario, in which the player is found inside a building, playing as a character who has been asked to deliver medical supplies to a friend. Shortly afterwards, a huge earthquake shall trap the building and its residents, introducing an 1https://www.dropbox.com/s/g4c872t9a6uwcfh/theDelivery.zip?dl=0 unexpected turn of events, while the player will have to deal with unsettling situations.
Throughout the game, the player is given the opportunity to
have conversations with in-game characters that lead to diferent
outcomes. Additionally, the player is able to interact with in-game
surroundings and objects. Even though many of the given choices
and actions may seem insignificant to the player, they have been
carefully devised to correspond to diferent criteria used in the
diagnosis of MDD. These criteria have been extracted from the
Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition
(DSM-V) [
More specifically, the conversations with in-game characters are implemented in such ways that, diferent player answers signal diferent probability (weight) of the player meeting a particular criterion of MDD. Moreover, with the help of a commonly used AI method in video games termed Behavior Trees, conversation flows are highly flexible and are adjusted suitably to maximize not only information extraction, but also the feeling of a realistic dialogue.
After gathering data from gameplay sessions of MDD and nonMDD players, we used Machine Learning methods to develop classiifers that can predict with a probability whether a gameplay session belongs to a player with MDD. 2.2
As in any serious game, the purpose is dual: entertainment for the player, and achievement of the specific "serious" goal. The first element includes all entertaining features, which are related to the player’s perspective about the game.
The second element includes the mechanisms that make this game a serious game. For this project, it was essential to have a good understanding of both dimensions.
In "The Delivery", the player has to look after his in-game statistics, i.e. Health, Fatigue and Sanity. For the player to keep these statistics within limits, he has to explore the environment, find hidden items or complete side-quests.
To achieve our goals, the player has to imagine himself within the scenario, in order to make the diagnosis more accurate, therefore the game was designed with this in mind. In detail, it belongs to a first-person video game category, so that the player can only imagine the character’s look, and the character’s voice is never heard.
The most important statistic within the game is the final score , which is an indication of the MDD level, according to the in-game mechanics. The player is not aware of the existence of this metric, so as not to become influenced by it.
The final score is computed in various ways. For instance, it changes every time the player interacts with an in-game character. Every answer during a discussion has a diferent weight relative to a particular diagnostic criterion extracted from DSM-V (e.g. the player admits that he/she does not sleep well at nights usually). When a given answer is related to a specific DSM-V criterion, the ifnal score changes accordingly.
Another way to increase the final score is through interaction with the environment. In order to keep statistics high, the player needs to interact with items within the surroundings. Showing sensitivity to characters and circumstances alters the final score in favor of a non-MDD label. 2.3
During a gameplay session, the player’s actions and their outcomes are recorded. This data includes: changes in the player’s basic statistics (Health, Fatigue, Sanity), the player’s answers during discussions (in chronological order), and the side quests successfully completed.
We collected 26 gameplay sessions from two groups: players
who were diagnosed with MDD (24%) and players who were not
diagnosed with MDD and were highly unlikely to sufer from MDD
(76%). The second group included cases who had never received
an oficial MDD diagnosis in the past, and were asked to fill in
the Patient Health Questionnaire (PHQ-9) [
Initially, we performed a visualization procedure in order to
visualize the game’s capability to distinguish MDD-positive and
MDDnegative behaviors within the game, using Principal Component
Analysis (PCA) [
Serious Game Development for the Diagnosis of Major Depressive Disorder
Then, we applied 4 classification methods: Decision Trees [
The RF classifier used = 20 trees and the kNN classifier used = 6 neighbors with the Euclidean distance metric. For the kNN classifier, we attempted to use an odd number of neighbors to avoid ties, but the optimal results were produced with = 6. 3
In this section we present visualization results using the PCA dimensionality reduction method and performance evaluation results of Machine Learning models trained using gameplay data. Principal Component Analysis was applied for = 2 and = 3 (Figure 2) components to the data. Visualization of the resulting data points indicates that the game design has employed in a relatively accurate manner diagnostic criteria for MDD. More particularly, in both experiments, results show a dense area of MDD-positive cases, which represent their in-game behavior, with only a few outlier cases. On the contrary, MDD-negative cases are more sparse and do not follow particular behavioral patterns, which is expected since the game mechanics do not target such behaviors.
Additionally, we trained various classification models based on the gameplay data, with the purpose of predicting whether a gameplay session belongs to an MDD-positive or MDD-negative player. Even with such a small, imbalanced dataset such ours, all models after training showed promising results in managing to distinguish gameplay sessions of MDD-positive and MDD-negative players.
More particularly, we measured the performance of each model
using the Accuracy, Precision, Recall, F1-score and F1-weighted
score metrics, computed using a Leave-One-Out Cross Validation
[
Random Decision Trees SVM (linear) kNN RF Accuracy 46.2% 65.4% 80.8% 77% 73.1% Precision 48.2% 53.7% 72.8% 38.5% 63.5% Recall 47.5% 54.2% 70% 50% 65% F1 43.1% 53.8% 71.2% 43.5% 64.1% F1-w 50.2% 66.3% 80.1% 66.9% 73.8% Table 1: Performance of diferent classifiers in predicting MDD cases from gameplay data. 4
In this paper, we developed a serious game prototype system called "The Delivery" for the diagnosis of Major Depressive Disorder (MDD). This video game, which features Artificial Intelligence methods, was designed with special care taken when implementing in-game mechanics, so as to correspond to oficial diagnostic criteria of the particular disorder. After recording gameplay sessions of players who were diagnosed with MDD and players who were highly unlikely to have MDD, we were able to train Machine Learning models that showed promising performance in distinguishing positive from negative MDD cases in gameplay behaviors.
Our proposed method is a novel concept, since no other work has touched the subject of MDD diagnosis in such perspective. It is crucial that new, more accurate and eficient techniques are developed, which are better suited to current and future societal needs. Additionally, this prototype can be further improved by extending the current scenario and gathering more data, which will lead to higher diagnostic accuracy of the system. A larger amount of gameplay data will also give us the opportunity to try other Machine Learning methods as well (e.g. Neural Networks). Moreover, this system can be adapted accordingly, for the purpose of identifying other types of depression, or mental health issues in general. Lastly, the video game industry is an ideal stakeholder for incorporating such diagnostic methodologies in video games and services.