2.1. Participants
Sixty right-handed, healthy volunteers (36 females, mean age ± SD: 25.71 ± 5.47 years) participated in the study after providing written informed consent. Participants were recruited through an online questionnaire. No participant had a history of neurologic, general medical or psychiatric conditions. The experimental protocol was approved by the Institutional Ethics Committee at the University G. d'Annunzio, Chieti-Pescara and it was performed in compliance with Declaration of Helsinki’s ethical principles. All participants were assessed for their schizotypal personality traits using the Schizotypal Personality Questionnaire 34.
2.1.2 Schizotypal Personality Questionnaire (SPQ)
The Schizotypal Personality Questionnaire34 is a validated questionnaire based on the diagnostic criteria of the DSM-III-R for schizotypal personality disorder. It consists of 74 items that measure the three-factor construct of schizotypy: cognitive-perceptual deficit, interpersonal deficit, and disorganization7,34. Responses are dichotomous (yes/no), with one point assigned for each agreement with an item. The total SPQ scores ranged from 0 to 53 (mean ± SD: 22.58 ± 13.37); the cognitive-perceptual scores ranged from 0 to 26 (mean ± SD: 9.56 ± 7.46); the interpersonal scores ranged from 0 to 29 (mean ± SD: 11.3 ± 6.82); the disorganization scores ranged from 0 to 16 (mean ± SD: 4.83 ± 4.08).
2.2. Procedure
Each participant completed the entire battery, which comprised tasks and questionnaires evaluating either exteroceptive or interoceptive bodily dimensions (Table 1).
2.2.1 Exteroceptive Bodily Dimensions
2.2.1.1 Body Image
Body Image refers to our subjective experience of the physical structure of our body in terms of its size, shape, and physical composition35. This dimension was evaluated through both a task, the Photographic Figure Rating Scale (PFRS) task, and a questionnaire, the Body Uneasiness Test (BUT).
The PFRS, adapted from the study by Naor-Ziv et al.36, was used to assess participants’ body image perception and body shape dissatisfaction. From this task, we quantified the following variables: the mean Body Mass Index (BMI) reflecting each participant's actual physique (A), ideal body image (I), and perception of how others perceive their body (O). Furthermore, we calculated the discrepancies between these means and the participants' actual BMI (R), denoted as ΔAR (A-R), ΔIR (I-R), and ΔOR (O-R). These differences provided insights into the participants' levels of misperception regarding their current physique, dissatisfaction with their ideal body image, and inaccurate beliefs about how others view their bodies. See Supplementary material for a detailed description of the task.
The Body Uneasiness Test is a questionnaire designed to assess individuals' attitudes towards their own body image. Developed by Cuzzolaro et al.37, the BUT38–40 consists of 34 items and a list of 37 body parts, characteristics, or functions. Participants rate each item on a scale ranging from "Never" (0) to "Always" (5), with higher scores indicating a greater level of impairment or uneasiness. The BUT calculates measures of overall severity and distress related to body uneasiness. These include the Global Severity Index (GSI), defined as BUT A score, and Positive Symptom Total (PST) defined as BUT B.
2.2.1.2 Spatial Tactile Acuity
Spatial Tactile Acuity refers to the ability to precisely perceive the location and quality of touch41. This dimension was evaluated through the two-point discrimination (2PD) task, which assesses the ability of participants to identify two closely spaced points on a small area of the skin and measures the accuracy of their discrimination skills42. We compared two body parts chest and thigh used as targets with the neck used as reference43. From this task, we quantified the following variables: the correct response for the two target body parts, chest (C-che) and thigh (C-thi); the overestimation for both chest (O-che) and thigh (O-thi) calculated by the errors and overestimation of the body parts compared to the reference and similarly, the underestimation for chest (U-che) and thigh (U-thi) calculated by errors and underestimation of the body parts. See Supplementary material for a detailed description of the task.
2.2.1.3 Body Structural Representation
Body Structural Representation (BSR) refers to knowledge about the topological organization of bodies, outlining how different body parts interrelate within a spatial configuration, focusing on the spatial positioning of each body part related to others35. This dimension was evaluated through the Finger Localization Task44, which requires participants to identify and differentiate stimulated fingers in three different conditions. We focused on the third condition in which participants had to identify which two fingers were simultaneously touched. We calculated the accuracy and included the score of correct responses (HIT) as a variable. See Supplementary material for a detailed description of the task.
2.2.1.4 Multisensory Integration
Multisensory Integration refers to the process by which inputs from two or more sensory modalities are combined by the nervous system to form a stable and coherent percept of the world 45. This dimension was evaluated through the Multisensory Integration (MSI) task aiming at investigating the integration of visual, auditory, and tactile stimuli in participants' perception. The task focused on evaluating their response speed in relation to the presentation of these stimuli46,47. For each multisensory pair – audio-visual (av), audio-tactile (at) and visuo-tactile (vt) - the analysis involved calculating the area-under-the-curve (AUC) subtended by the distribution of reaction times to multisensory stimuli as compared to the distribution of reaction times to unimodal stimuli. Hence, AUC-av, AUC-at and AUC-vt represent a proxy of the magnitude of multisensory integration for audio-visual, audio-tactile and visuo-tactile stimuli, respectively48. See Supplementary material for a detailed description of the task.
2.2.1.5 Multisensory Temporal Resolution
Multisensory Temporal Resolution refers to the principle that optimal multisensory integration occurs when stimuli from different senses are presented closely in time, diminishing as the temporal gap increases. This principle highlights the critical role of temporal proximity in influencing the nervous system's integration of diverse sensory information49.
This dimension was evaluated through the Simultaneity Judgment (SJ) task, which aims to measure temporal sensitivity in the integration of multisensory stimuli. Specifically, we focused on auditory and tactile stimuli (A-T). Two parameters are usually derived from this task: the point of subjective equality (PSE), providing an estimate of the interval between stimuli at which there is the highest probability of the perception of simultaneity and the ‘just noticeable difference’ (JND-sj), reflecting the subject’s sensitivity to changes in temporal intervals between the stimuli. The JND value denotes the minimal temporal interval at which the change between the perceived temporal relation stimuli can be observed50. See Supplementary material for a detailed description of the task.
2.2.1.6 Peripersonal Space
Peripersonal Space refers to the space surrounding the body where the integration of stimuli on the body and from the external environment is facilitated51.
This dimension was evaluated through the Peripersonal Space (PPS) task9,52,53, which allows evaluating the individual’s boundaries of the peripersonal space by assessing the optimal temporal interval for the integration of tactile and auditory stimuli54,55. To this aim, a psychometric function is fitted to the RT data. The PSE of the psychometric function is taken as a proxy of the peripersonal space boundary. See Supplementary material for a detailed description of the task.
2.2.1.7 Temporal Tactile Acuity
Temporal Tactile Acuity refers to the ability to perceive and discriminate temporal aspects of tactile stimuli. It involves the capacity to detect and distinguish temporal characteristics of sensations related to touch56. This dimension was evaluated through the Temporal Order Judgment. In a typical Temporal-Order Judgment (TOJ) task, participants are tasked with determining the order of two tactile stimuli presented sequentially to their left and right hands. When the hands are in an “uncrossed” position (toju), participants can rely on tactile and proprioceptive cues related to their body posture, utilizing a body-centred reference frame57. From this task, we quantified the JND (JND-toju), as a measure of precision, and the proportion of correct responses58. The JND-toju represents the smallest interval at which the participants can reliably decide which sensory input of the two presented was first59. See Supplementary material for a detailed description of the task.
2.2.1.8 Sensorimotor functions
Touch remapping
To perceive the location of touch in space, the brain combines information about touched skin location with information about the location of that body part in space. When the two hands are crossed, this integration is impaired, affecting the ability to judge the order of touches on both hands58. This crossed-hand posture creates a conflict between how tactile senses represent external space. Consequently, the same cues must be remapped using an external reference frame. This remapping process becomes necessary to accurately judge the temporal order of the tactile stimuli in the crossed-hand condition (tojc)17.
From this task, we measured the following variables: the JND (JND-tojc), which represents the smallest interval at which participants can reliably determine which of the two presented sensory inputs came first 59 and the Sum of Confusions (SC-toj), that indicates the sum of differences in the response functions between crossed and uncrossed conditions17. SC-toj is a global indicator of differences, which provides an overarching measure of the divergence between the two conditions and assess increases in judgment reversals resulting from the arm-crossing manipulation60. See Supplementary material for a detailed description of the task.
Laterality Judgment Task
The Laterality Judgement Task (LJT) is designed to assess participants' ability to mentally rotate the presented hand images and accurately judge the lateral orientation of presented hand images61. We calculated slopes, which reflect the efficiency of the neural mechanism underlying the mental rotation process: a smaller slope indicates higher neural efficiency in mental rotation62. We quantified slopes for both hands, indicated as Mental Rotation Efficiency for the left hand (MRE-lh) and the right hand (MRE-rh), illustrating the efficiency of the respective mental rotation processes63.
2.2.2 Interoceptive Bodily Dimensions
2.2.2.1 Interoceptive Accuracy
Interoceptive accuracy refers to the accuracy in detecting internal bodily sensations29. Two tasks determined cardiac interoceptive accuracy: the Heartbeat Detection task, also known as the Tapping or Tracking task64,65, and the Heartbeat Counting task66. In the Heartbeat Detection task (d) participants were instructed to focus their attention on their bodily sensations and to press a button on a device each time they felt a heartbeat. To calculate the accuracy during the Heartbeat Detection task (Acc-d), the recorded R-peaks and the corresponding tapping times were compared67–69. The Heartbeat Counting task (c) measures cardiac interoceptive accuracy by evaluating participants' performance in silently counting their felt heartbeats during specific time intervals 29. Following each time interval, participants were required to verbally report the count or estimated number of heartbeats they perceived. Next, the accuracy score (Acc-c) was calculated by comparing the recorded heartbeats and the counted heartbeats70. See Supplementary material for a detailed description of the task.
2.2.2.2 Interoceptive Sensibility
Interoceptive Sensibility is the subjective account of experiencing internal bodily sensations30. This dimension can be assessed using subjective measures that index both the individual's confidence in their interoceptive ability and their interoceptive feelings29. We evaluated this dimension through the score of confidence in interoceptive accuracy during the performance of the Heartbeat Detection task (Con-d) and the Heartbeat Counting task (Con-c). This confidence rating was assessed using a scale ranging from 1 to 9, with 1 indicating low confidence (total guess/no heartbeat awareness) and 9 indicating high confidence (complete perception of heartbeat). We also used two self-report questionnaires: the Body Perception Questionnaire (BPQ)71,72 and the Multidimensional Assessment of Interoceptive Awareness (MAIA)73. The BPQ generally evaluates body awareness and autonomic symptoms. From the BPQ we obtained a score for each subscale: the body awareness (BOA) that consists of items related to the upper parts of the body, the supradiaphragmatic (SUP) that is involved in regulating the functions of organs situated above the diaphragm and additionally, the subdiaphragmatic/body awareness factor (BOA/SUB), includes items related to subdiaphragmatic issues. The MAIA generally evaluates multiple dimensions of interoception. From the MAIA we obtained a score for each subscale: Noticing (1M), Not-Distracting (2M) Not-Worrying (3M) Attention Regulation (4M) Emotional Awareness (5M) Self-Regulation (6M) Body Listening (7M) Trusting (8M). See supplementary material for a detailed description of the task and questionnaires.
2.2.2.3. Interoceptive Awareness
Interoceptive Awareness is the metacognitive awareness of interoceptive accuracy and refers to the correspondence between objective interoceptive accuracy and subjective confidence report29,74. We quantified the participant’s awareness by comparing the accuracy and the sensibility in both the Heartbeat Detection task (Aw-d) and the Heartbeat Counting task (Aw-c).
2.3. Statistical Analysis: Machine Learning Approach
When attempting to predict an output based on input that may be correlated with each other, the high collinearity among features, combined with the high number of features compared to the number of samples (such as subjects or tests in our case), can potentially disrupt the results, leading to unstable predictions that are sensitive to noise and susceptible to overfitting and poor generalization. Various linear and nonlinear regression and classification algorithms have been created to avoid the issue of overfitting. These methods can employ different approaches, including penalizing the fitting parameter through techniques like regularization or reducing the dimensionality of the feature space to mitigate overfitting. In this study, a linear regression analysis was conducted using a technique that involves reducing the dimensionality of the feature space, namely the Partial Least Square Regression (PLSR). PLSR has been widely demonstrated to be successful in mitigating overfitting, especially when dealing with collinearity. PLSR is based on the fundamental assumption that the observed data is produced by a system or process influenced by a limited number of latent variables that are not directly observed or measured. PLSR enables the creation of regression equations by condensing the predictors into a more concise set of uncorrelated components, which are linear combinations of the initial predictors. This approach conducts regression on these components, identifying those that capture the most pertinent information within the independent variables. This aids in predicting the dependent variable while simultaneously reducing the complexity of the regression problem by employing a smaller number of components than the original count of independent variables. PLSR can be viewed as a supervised learning method, a sort of supervised Principal Component Analysis. In contrast to PCA, which reduces the dimensionality of the feature space by analysing the eigen solutions of the covariance matrix in an unsupervised manner, PLSR is a supervised learning algorithm. PLSR achieves this by determining feature space components that maximize the covariance between the independent and dependent variables, making it another form of supervised learning algorithm. The PLSR algorithm necessitates an a priori selection of one parameter, known as a hyperparameter, i.e., the number of uncorrelated components (K) to be used for regression. Generally, to conduct hyperparameter optimization, a training set, a validation set, and a test set are required. The training set is used to train the algorithm (e.g., estimate PLSR weights), the validation set is used to optimize the hyperparameter (e.g., estimate the optimal K), and the test set is used to assess the generalization performance of the learning process. This separation of data implies a reduced sample size as part of a trade-off choice among the sets.
Furthermore, another approach that minimizes the loss of samples in the different sets while evaluating the algorithm's generalization capabilities is cross-validation (CV). In CV, the data is divided into folds, and the model is trained on all data except one-fold in an iterative manner. The out-of-sample performance (i.e., generalization) is assessed based on the remaining fold and averaged across iterations. If the number of folds equals the number of samples, which is often referred to as "leave-one-out" cross-validation, it provides a rigorous assessment of the model's generalization performance. If the hyperparameters of the model need to be optimized, the optimization process cannot be reliably determined based on a simple CV without sacrificing a generalizable estimate of model performance. CV can be adapted to simultaneously select the best set of hyperparameters and provide a generalizable error estimation through a procedure called nested CV. Starting with k folds, nested cross-validation (nCV) is performed with an outer loop of k folds and an inner loop of k − 1 folds. The outer loop estimates the generalization performance of the model (i.e., test), while the inner loop evaluates the optimal hyperparameters (i.e., validation).
In each iteration, a fold is selected as the outer set (for assessing generalization), and the remaining k − 1 sets are combined to form the corresponding outer training set. Then, each outer training set is further subdivided into j sets, and one set is iteratively chosen as the inner test set (for evaluating the optimal hyperparameters), with the j − 1 other sets forming the corresponding inner training set.
If the number of folds equals the number of samples for both loops, the procedure is referred to as leave-one-out nCV. In this study, a leave-one-out nCV was implemented to assess the PLSR generalization performance and the optimal number of components using the following variables as the dependent variables: SPQ-CP, SPQ-I, SPQ-D and SPQ-TOT. For each of them, PLSR was used to test three models, each of which used exteroceptive, interoceptive, and intero-exteroceptive variables (BMI (R), A (actual), I (ideal), O (other), ΔAR, ΔIR, ΔOR, Pt. tot. A, Pt. tot. B, C-che, O-che, U-che, U-thi, O-thi, C-thi, HIT, AUC-av, AUC-at, AUC-vt, PSE-sj, JND-sj, PSE-pps, JND-toju, JND-tojc, SC-toj MRE-lh, MRE-rh, Acc-d, Acc-c, Con-d, Con-c, BOA, SUP, BOA/SUB, 1M-8M, Aw-d, Aw-c) as independent variables, described in table 1, resulting in a total of 12 models. Each model was trained and tested separately.
The regression performances of the models were evaluated through correlation analyses between the predicted values and the actual values (which likely stand for some specific variable or outcome). The correlation coefficient, denoted as "r," is reported for each tested model. The degrees of freedom (DF) and the significance of the null hypothesis (p) are also reported in the analysis. These values help assess the strength and significance of the correlations between the predicted and actual values.
Furthermore, the statistical significance of the weights was assessed using a randomization procedure, which involved constructing a confidence interval for the null hypothesis of each weight being zero (iterating the same algorithm 1 0^6 times on random shuffled outcomes).