Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Multi-modal automated major depressive disorder detection
Thesis (PhD)--Stellenbosch University, 2025.
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Published: |
Stellenbosch : Stellenbosch University
2026
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867613953811546112 |
|---|---|
| access_status_str | Open Access |
| author | Shumba, Sandura |
| author2 | Coetzer, Hanno |
| author_browse | Coetzer, Hanno Shumba, Sandura |
| author_facet | Coetzer, Hanno Shumba, Sandura |
| author_sort | Shumba, Sandura |
| collection | Thesis |
| dc_rights_str_mv | Stellenbosch University |
| description | Thesis (PhD)--Stellenbosch University, 2025. |
| format | Thesis |
| id | oai:scholar.sun.ac.za:10019.1/134818 |
| institution | Stellenbosch University (South Africa) |
| last_indexed | 2026-06-10T12:44:20.637Z |
| license_str | Other — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from SUNScholar — Stellenbosch University Repository |
| publishDate | 2026 |
| publishDateRange | 2026 |
| publishDateSort | 2026 |
| publisher | Stellenbosch : Stellenbosch University |
| publisherStr | Stellenbosch : Stellenbosch University |
| record_format | dspace |
| source_str | SUNScholar — Stellenbosch University Repository |
| spelling | oai:scholar.sun.ac.za:10019.1/134818 Multi-modal automated major depressive disorder detection Shumba, Sandura Coetzer, Hanno Stellenbosch University. Faculty of Science. Dept. of Mathematical Sciences. Applied Mathematics Division. Combined modality therapy Depression, Mental -- Diagnosis Biomedical engineering Deep learning (Machine learning) Thesis (PhD)--Stellenbosch University, 2025. Shumba, S. 2025. Multi-Modal Automated Major Depressive Disorder Detection. Unpublished doctoral dissertation. Stellenbosch: Stellenbosch University [online]. Available: https://scholar.sun.ac.za/items/8d1847ce-a5a1-49a9-a422-e4dd5e7d34fc ENGLISH ABSTRACT: Major Depressive Disorder (MDD) presents a substantial global health challenge, necessitating diagnostic tools capable of robustly capturing neural dynamics and emotional expressions. This dissertation introduces a multi-modal framework that integrates generalised Partial Directed Coherence (gPDC) measures derived from EEG data and Log-Based Mel Spectrum (LBMS) images extracted from audio data. The model features two specialised encoders, enhanced with Convolutional Block Attention Mechanisms (CBAMs), to independently extract diagnostically relevant neurophysiological and acoustic features. These outputs are fused via a cross-modal attention mechanism, which captures inter-modality dependencies for more effective feature integration. A key advancement of this work is the adoption of a rigorous patientcentric data splitting approach, which ensures complete participant separation across training, validation, and test phases. This method also guarantees that data from multiple modalities belonging to the same patient are processed simultaneously, preserving the natural correlation between EEG and audio features. This methodology eliminates the risk of data leakage, guaranteeing reliable generalisation—a limitation in many prior studies that rely on random splits and report inflated a ccuracies. T he p roposed f ramework a chieves an accuracy of 97.86%, the highest among patient-centric approaches for MDD detection, delivering a robust and clinically reliable diagnostic system that advances the state-of-the-art in multi-modal depression detection. AFRIKAANSE OPSOMMING: Ernstige Depressiewe Versteuring (EDV) stel ’n wesenlike wêreldwye gesondheidsuitdaging, wat diagnostiese instrumente noodsaak wat in staat is om neurale dinamika en emosionele uitdrukkings sterk vas te lê. Hierdie proefskrif stel ’n multi-modale raamwerk bekend wat Veralgemeende Gedeeltelike Gerigte Koherensie (VGGK) maatreëls, afgelei van EEG-data en log-gebaseerde Mel Spectrum (LBMS) beelde wat uit oudio-data onttrek is integreer. Die model beskik oor twee gespesialiseerde enkodeerders, verbeter met Konvolusie Blok Aandag-Meganisme (KBAM), om onafhanklik diagnosties relevante neurofisiologiese en akoestiese kenmerke te onttrek. Hierdie uitsette word saamgesmelt via ’n kruismodale aandagmeganisme, wat intermodaliteitsafhanklikhede vaslê vir meer effektiewe kenmerkintegrasie. ’n Sleutelvordering van hierdie werk is die aanvaarding van ’n streng pasiëntgesentreerde dataverdelingsbenadering, wat volledige deelnemerskeiding oor opleiding-, validerings- en toetsfases verseker. Hierdie metode waarborg ook dat data van verskeie modaliteite wat aan dieselfde pasiënt behoort gelyktydig verwerk word, wat die natuurlike korrelasie tussen EEG en oudiokenmerke behou. Hierdie metodologie elimineer die risiko van datalekkasie, wat betroubare veralgemening waarborg —’n beperking in baie vorige studies wat staatmaak op ewekansige verdelings en ooroptimistiese akkuraatheid rapporteer. Die voorgestelde raamwerk bereik ’n akkuraatheid van 97.86%, die hoogste onder pasiënt-gesentreerde benaderings vir MDD-opsporing, wat ’n robuuste en klinies betroubare diagnostiese stelsel lewer wat die staat-van-die-kuns in multi-modale depressie-opsporing bevorder. Doctoral 2026-01-09T08:44:06Z 2026-01-09T08:44:06Z 2025-12 Thesis https://scholar.sun.ac.za/handle/10019.1/134818 Stellenbosch University xviii, 136 pages : illustrations application/pdf Stellenbosch : Stellenbosch University |
| spellingShingle | Combined modality therapy Depression, Mental -- Diagnosis Biomedical engineering Deep learning (Machine learning) Shumba, Sandura Multi-modal automated major depressive disorder detection |
| title | Multi-modal automated major depressive disorder detection |
| title_full | Multi-modal automated major depressive disorder detection |
| title_fullStr | Multi-modal automated major depressive disorder detection |
| title_full_unstemmed | Multi-modal automated major depressive disorder detection |
| title_short | Multi-modal automated major depressive disorder detection |
| title_sort | multi modal automated major depressive disorder detection |
| topic | Combined modality therapy Depression, Mental -- Diagnosis Biomedical engineering Deep learning (Machine learning) |
| url | https://scholar.sun.ac.za/handle/10019.1/134818 |
| work_keys_str_mv | AT shumbasandura multimodalautomatedmajordepressivedisorderdetection |