Neural Entropy Pattern : An EEG Representation for Predicting Depression Therapy Outcomes

We propose the Neural Entropy Pattern (NEP), an electroencephalography (EEG) representation intended to separate responders from non-responders before treatment. NEP consists of ten entropy measures, computed from 19 channels in five frequency bands and the broadband signal. It was used as a $20 \times 57$ image by twelve deep learning models and as a flat vector by seven shallow classifiers, and was evaluated on a selective serotonin reuptake inhibitor (SSRI) dataset ($n = 30$) and a repetitive transcranial magnetic stimulation (rTMS) dataset ($n = 61$) under leave-one-subject-out (LOSO) and patient-level five-fold cross-validation, with one prediction per patient and without using the held-out patients for training, early stopping or model selection. The highest balanced accuracy (BA) was, for SSRI, 76.4\% under LOSO (EfficientNet-B0; area under the receiver operating characteristic curve, AUC, 0.89) and 72.2\% under five-fold cross-validation (linear support vector machine, SVM; AUC 0.77) and, for rTMS, 62.1\% ($k$-nearest neighbours; AUC 0.62) and 63.7\% (linear SVM; AUC 0.63). These are the best of 19 models, selected post hoc (95\% confidence intervals 23--32 points). Both SSRI results were above chance by AUC (Holm-adjusted $p = 0.001$ and 0.027). Log-energy and differential entropy carried most of the Shapley-value importance (72\% for SSRI, 60\% for rTMS). For SSRI, the largest group differences were at the left fronto-temporal electrode F7 ($\delta$, $\theta$, $\beta$ and broadband domains; relatively higher power in responders), and the most important features were at F7 and T3 ($\beta$ and $\gamma$ rhythms, broadband signal); for rTMS, the most important features were at the right frontal electrode F4 ($\alpha$ rhythm, broadband signal). The NEP image performed comparably to a multichannel continuous wavelet transform image, and the NEP vector to statistical features.

Authors

Institutions

Publication Details

Journal
Biomedical Physics & Engineering Express
Published
2026-11-19
DOI
https://doi.org/10.1088/2057-1976/aec455
Primary Topic
EEG and Brain-Computer Interfaces
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Neural Entropy Pattern : An EEG Representation for Predicting Depression Therapy Outcomes

Muhammad Tariq Sadiq, Mutlu Mete, Siuly Siuly, Hesam Akbari et al.
Biomedical Physics & Engineering Express
EEG and Brain-Computer Interfaces
article

Neural Entropy Pattern : An EEG Representation for Predicting Depression Therapy Outcomes

Muhammad Tariq Sadiq, Mutlu Mete, Siuly Siuly, Hesam Akbari, Niloofar Alizadeh, Reza Kazemi, Hana Rostami, Tayyaba Shahwar, Hamed Shamsi
article en

Abstract

We propose the Neural Entropy Pattern (NEP), an electroencephalography (EEG) representation intended to separate responders from non-responders before treatment. NEP consists of ten entropy measures, computed from 19 channels in five frequency bands and the broadband signal. It was used as a $20 \times 57$ image by twelve deep learning models and as a flat vector by seven shallow classifiers, and was evaluated on a selective serotonin reuptake inhibitor (SSRI) dataset ($n = 30$) and a repetitive transcranial magnetic stimulation (rTMS) dataset ($n = 61$) under leave-one-subject-out (LOSO) and patient-level five-fold cross-validation, with one prediction per patient and without using the held-out patients for training, early stopping or model selection. The highest balanced accuracy (BA) was, for SSRI, 76.4\% under LOSO (EfficientNet-B0; area under the receiver operating characteristic curve, AUC, 0.89) and 72.2\% under five-fold cross-validation (linear support vector machine, SVM; AUC 0.77) and, for rTMS, 62.1\% ($k$-nearest neighbours; AUC 0.62) and 63.7\% (linear SVM; AUC 0.63). These are the best of 19 models, selected post hoc (95\% confidence intervals 23--32 points). Both SSRI results were above chance by AUC (Holm-adjusted $p = 0.001$ and 0.027). Log-energy and differential entropy carried most of the Shapley-value importance (72\% for SSRI, 60\% for rTMS). For SSRI, the largest group differences were at the left fronto-temporal electrode F7 ($\delta$, $\theta$, $\beta$ and broadband domains; relatively higher power in responders), and the most important features were at F7 and T3 ($\beta$ and $\gamma$ rhythms, broadband signal); for rTMS, the most important features were at the right frontal electrode F4 ($\alpha$ rhythm, broadband signal). The NEP image performed comparably to a multichannel continuous wavelet transform image, and the NEP vector to statistical features.

Biomedical Physics & Engineering Express
University of Essex (GB), University of North Texas (US), Harvard University (US), Karadeniz Technical University (TR), University of Tehran (IR), Arizona State University (US), Victoria University (AU)
EEG and Brain-Computer Interfaces
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon