Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
A deep learning system for x-ray analysis and patient diagnosis
Thesis (MEng)--Stellenbosch University, 2023.
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | en_ZA |
| Published: |
Stellenbosch : Stellenbosch University
2023
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867613873102651392 |
|---|---|
| access_status_str | Open Access |
| author | Faure, James |
| author2 | Engelbrecht, Andries |
| author_browse | Engelbrecht, Andries Faure, James |
| author_facet | Engelbrecht, Andries Faure, James |
| author_sort | Faure, James |
| collection | Thesis |
| dc_rights_str_mv | Stellenbosch University |
| description | Thesis (MEng)--Stellenbosch University, 2023. |
| format | Thesis |
| id | oai:scholar.sun.ac.za:10019.1/128685 |
| institution | Stellenbosch University (South Africa) |
| language | en_ZA |
| last_indexed | 2026-06-10T12:43:03.527Z |
| license_str | Other — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from SUNScholar — Stellenbosch University Repository |
| publishDate | 2023 |
| publishDateRange | 2023 |
| publishDateSort | 2023 |
| publisher | Stellenbosch : Stellenbosch University |
| publisherStr | Stellenbosch : Stellenbosch University |
| record_format | dspace |
| source_str | SUNScholar — Stellenbosch University Repository |
| spelling | oai:scholar.sun.ac.za:10019.1/128685 A deep learning system for x-ray analysis and patient diagnosis Faure, James Engelbrecht, Andries Stellenbosch University. Faculty of Engineering. Dept. of Industrial Engineering. Deep learning (Machine learning) Computer vision in medicine Orthodontics, Corrective Teeth -- Radiography Thesis (MEng)--Stellenbosch University, 2023. ENGLISH ABSTRACT: In South African public health, orthodontic radiology has de_ciencies in both capacity and expertise. A patient arrives at a hospital to receive treatment for oral anomalies and diseases. After waiting in a long queue, the patient receives an X-ray of their oral area. Between screening and a clinic doctor, a maxillofacial radiologist is tasked to perform a diagnostic procedure where oral anomalies and diseases are identi_ed. A report is then sent to the dentist so the patient can be prescribed treatment. In South Africa, the radiologist is overloaded with too many X-rays to be analysed. A privation of diagnosis or misdiagnosis is often fatal for the patient. The goal of the thesis is demonstrate medical imaging diagnosis, using a dental X-ray dataset. Deep learning is a sub_eld of arti_cial intelligence that uses complex architectures to solve problems that have large problems. Computer vision is a process of deriving information from visual inputs. Deep learning has accelerated computer vision by analysing image datasets to identify predictable features. The number of class labels that a deep learning model can predict is dependent on the number of images that are annotated. Annotation of large datasets is timeconsuming and complex, especially in the cases where domain experts are needed. In the case of dentistry, over 50 tooth anomalies and hundreds of diseases are detectable in the oral area. To create one model to predict all the classes would be infeasible from a human resource perspective due to the extensive labelling process. It is more e_cient to train a variety of smaller models with each model specialized in a limited number of related diseases or anomalies. The models are then engineered into a multi-component system that is used to successfully diagnose a patient. The research _eld is referred to as machine learning engineering, where development is focused on the system as a whole rather than deep learning model architecture alone. The researcher must use engineering principles to account for hardware, software development, machine learning model training and deployment, and computer vision techniques. The results of the thesis provide evidence that deep learning can successfully be used to make predictions on dental X-rays. The problem of scarce data is solved by grouping a minimum number of classes per model and hence training several di_erent models. The models are then deployed in series and parallel using a queuing system for communication. The trained deep learning models are deployed into a highly scalable system that can successfully diagnose an X-ray in a quicker time than a radiologist. AFRIKAANSE OPSOMMING: Ortodontiese radiologie in Suid-Afrikaanse openbare gesondheid het tekortkominge ten opsigte van beide kapasiteit en kundigheid. 'n Pasi ent meld by 'n hospitaal aan om behandeling vir orale anomalie e en siektes te kry. Nadat hy/sy in 'n lang tou gewag het, ontvang die pasi ent 'n X-straal van hul mond. Tussen sifting en 'n kliniekdokter word 'n kaak-gesig-en-mond-radioloog getaak om 'n diagnostiese prosedure uit te voer waar orale afwykings en siektes ge _denti_seer is. 'n Verslag word dan aan die tandarts gestuur sodat daar behandeling vir die pasi ent voorgeskryf kan word. In Suid-Afrika is die radioloog _of orrlaai _of nie beskikbaar nie. 'n Gebrek aan diagnose of verkeerde diagnose is dikwels dodelik vir die pasi ent. Diepleer is 'n subveld van kunsmatige intelligensie wat komplekse argitekture gebruik om probleme met groot datastelle op te los. Rekenaarvisie is 'n proses om inligting uit visuele insette af te lei. Diepleer het die toepassing van rekenaarvisie versnel deur beelddatastelle te ontleed om voorspelbare kenmerke te identi_seer. Die aantal kategorie e wat 'n rekenaarvisie-diepleermodel kan voorspel, is afhanklik van die aantal beelde wat geannoteer is. Die annotering van groot datastelle is kompleks en tydrowend. In die geval van tandheelkunde is meer as 50 tandafwykings en honderde siektes in die mond waarneembaar. Om een model te skep om al hierdie werk te doen is onuitvoerbaar vanuit 'n menslikehulpbron-perspektief. Hierdie probleem word opgelos deur 'n verskeidenheid kleiner modelle op te lei wat binne 'n stelselargitektuur ontplooi word om die kommunikasie tussen verskeie modelle te fasiliteer sodat 'n pasi ent suksesvol gediagnoseer word. Elke model is 'n kenner in vier of minder tandheelkundige abnormaliteite. Die resultate van die proefskrif verskaf bewyse dat diepleer suksesvol gebruik kan word om voorwerpe op X-strale te identi_seer. Die probleem van skaars data word opgelos deur nie meer as vier kenmerke per model te groepeer nie en verskeie modelle op te lei. Die modelle word dan parallel ontplooi deur 'n toustaan-stelsel vir kommunikasie te gebruik. Die opgeleide diepleermodelle word in 'n hoogs skaalbare stelsel ontplooi waar 'n X-straal 'n suksesvolle diagnose vinniger as 'n radioloog kan lewer. Masters 2023-02-13T13:29:28Z 2023-11-16T08:13:39Z 2023-02-13T13:29:28Z 2023-11-16T08:13:39Z 2023-02 Thesis https://scholar.sun.ac.za/handle/10019.1/128685 en_ZA Stellenbosch University xxx, 228 pages : illustrations application/pdf Stellenbosch : Stellenbosch University |
| spellingShingle | Deep learning (Machine learning) Computer vision in medicine Orthodontics, Corrective Teeth -- Radiography Faure, James A deep learning system for x-ray analysis and patient diagnosis |
| title | A deep learning system for x-ray analysis and patient diagnosis |
| title_full | A deep learning system for x-ray analysis and patient diagnosis |
| title_fullStr | A deep learning system for x-ray analysis and patient diagnosis |
| title_full_unstemmed | A deep learning system for x-ray analysis and patient diagnosis |
| title_short | A deep learning system for x-ray analysis and patient diagnosis |
| title_sort | deep learning system for x ray analysis and patient diagnosis |
| topic | Deep learning (Machine learning) Computer vision in medicine Orthodontics, Corrective Teeth -- Radiography |
| url | https://scholar.sun.ac.za/handle/10019.1/128685 |
| work_keys_str_mv | AT faurejames adeeplearningsystemforxrayanalysisandpatientdiagnosis AT faurejames deeplearningsystemforxrayanalysisandpatientdiagnosis |