HZHadi Zaatiti

Hadi Zaatiti

*This is an algorithmically generated author record based on publicly available information.

Hadi Zaatiti has a total of 9 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Sensor Array Geometry Acquisition by Spherical Coil Array for OPM-Based MEG System

    by Yoshiaki Adachi, Daisuke Oyama, Gen Uehara, Hadi Zaatiti

    Precise determination of sensor array geometry is critical for accurate magnetic source localization in magnetoencephalography (MEG). This is particularly important for optically pumped atomic magnetometer (OPM)-based MEG systems, since their sensor array configuration can be flexibly modified depending on the subject’s head shape. In this study, we demonstrate, for the first time, that a spherical coil array—previously validated for SQUID-based MEG—can also be applied to OPM-based MEG systems. Using a dry-type MEG phantom equipped with 49 equivalent current dipoles (ECDs), we evaluated the sensor array geometry obtained with the spherical coil array. We compared it with that derived from the method originally implemented in the OPM-MEG system. To isolate geometry-related effects, we incorpolated two procedures: template fitting for field extraction and a recursive coordinate system transformation. These procedures minimized uncertainties not associated with the sensor array geometry in the ECD estimation. The results show that spherical coil calibration reduced the variability in the positional displacement between estimated and assumed ECDs in close (on-scalp) recordings by approximately 50%, as confirmed by 95% credible intervals. In the far mode, improvements were present but less pronounced. These findings indicate that the spherical coil array calibration enhances OPM-MEG source estimation by providing more accurate orientations and sensitivities. As a result, OPM-MEG measurements become more reliable, enabling broader applications in research and clinical settings.

  • Open Access

    Towards stable AI systems for Evaluating Arabic Pronunciations

    by Hadi Zaatiti, Hatem Hajri, Osama Abdullah, Nader Masmoudi

    Modern Arabic ASR systems such as wav2vec 2.0 excel at word- and sentence-level transcription, yet struggle to classify isolated letters. In this study, we show that this phoneme-level task, crucial for language learning, speech therapy, and phonetic research, is challenging because isolated letters lack co-articulatory cues, provide no lexical context, and last only a few hundred milliseconds. Recogniser systems must therefore rely solely on variable acoustic cues, a difficulty heightened by Arabic's emphatic (pharyngealized) consonants and other sounds with no close analogues in many languages. This study introduces a diverse, diacritised corpus of isolated Arabic letters and demonstrates that state-of-the-art wav2vec 2.0 models achieve only 35% accuracy on it. Training a lightweight neural network on wav2vec embeddings raises performance to 65%. However, adding a small amplitude perturbation (epsilon = 0.05) cuts accuracy to 32%. To restore robustness, we apply adversarial training, limiting the noisy-speech drop to 9% while preserving clean-speech accuracy. We detail the corpus, training pipeline, and evaluation protocol, and release, on demand, data and code for reproducibility. Finally, we outline future work extending these methods to word- and sentence-level frameworks, where precise letter pronunciation remains critical.

  • Open Access

    Phantom-Based Approach for Comparing Conventional and Optically Pumped Magnetometer Magnetoencephalography Systems

    by Daisuke Oyama, Hadi Zaatiti

    Magnetoencephalography (MEG) is a vital tool for understanding neural dynamics, offering a noninvasive technique for measuring subtle magnetic field variations around the scalp generated by synchronized neuronal activity. Two prominent sensor technologies exist: the well-established superconducting quantum interference device (SQUID) and the more recent optically pumped magnetometer (OPM). Although many studies have compared these technologies using human-subject data in neuroscience and clinical studies, a direct hardware-level comparison using dry phantoms remains unexplored. This study presents a framework for comparing SQUID- with OPM-MEG systems in a controlled environment using a dry phantom that emulates neuronal activity, allowing strict control over physiological artifacts. Data were obtained from SQUID and OPM systems within the same shielded room, ensuring consistent environmental noise control and shielding conditions. Positioning the OPM sensors closer to the signal source resulted in a signal amplitude approximately 3–4 times larger than that detected by the SQUID-MEG system. However, the source localization error of the OPM-MEG system was approximately three times larger than that obtained by the SQUID-MEG system. The cause of the large source localization error was discussed in terms of sensor-to-source distance, sensor count, signal–noise ratio, and the spatial coverage provided by the sensor array of the source signal.


  • A hyperbolic approach for learning communities on graphs

    by Thomas Gerald, Hadi Zaatiti, Hatem Hajri, Nicolas Baskiotis, Olivier Schwander