An automatic method for assessing spiking of tibial tubercles associated with knee osteoarthritis
Abstract
Polvinivelrikon kasvavan esiintyvyyden vuoksi tehokkaat varhaiset diagnoosimenetelmät ovat haluttavia. Radiografia on keskeinen osa polvinivelrikon diagnostiikassa. Polvinivelrikon varhainen tunnistaminen on haastavaa, sillä tärkeimpiä polvinivelrikon merkkejä on vaikea havaita röntgenkuvista taudin varhaisessa vaiheessa. Koneoppimallien kehittämistä varhaiseen polvinivelrikon tunnistamiseen vaikeuttaa lisäksi saatavilla olevan datan kohinaisuus. Tämän tutkielman tavoitteena oli tarkastella hypoteesia eminentian terävöitymisestä varhaisen polvinivelrikon piirteenä. Tutkielmassa kehitettiin myös neuroverkkopohjainen malli piirteen tunnistamiseen röntgenkuvista. Työn tulokset viittaavat eminentian terävyyden olevan yhteydessä varhaiseen polvinivelrikkoon. Tämän lisäksi piirre voidaan tunnistaa automaattisesti röntgenkuvista. Työn tuloksia voidaan pitää kuitenkin vasta alustavina.
Efficient and scalable early diagnostic methods are warranted due to the rising prevalence of knee osteoarthritis. Radiographic imaging is the standard procedure in osteoarthritis diagnosis. However, the circumstances for early diagnosis are problematic since the plain radiographs are insensitive to the established early signs of knee osteoarthritis. Furthermore, developing machine learning tools for radiographic knee osteoarthritis diagnosis is challenging due to noisy ground-truth. The objective of this thesis was to assess a feature called spiking of tibial tubercles, which has been hypothesized as an early sign of knee osteoarthritis. Additionally, we developed a model based on neural networks for identifying the feature in plain radiographs. Our results indicate promise in including tibial spiking as an early feature of knee osteoarthritis, and the feature is identifiable automatically. However, the work in the current thesis is limited and should be validated by future work.
Efficient and scalable early diagnostic methods are warranted due to the rising prevalence of knee osteoarthritis. Radiographic imaging is the standard procedure in osteoarthritis diagnosis. However, the circumstances for early diagnosis are problematic since the plain radiographs are insensitive to the established early signs of knee osteoarthritis. Furthermore, developing machine learning tools for radiographic knee osteoarthritis diagnosis is challenging due to noisy ground-truth. The objective of this thesis was to assess a feature called spiking of tibial tubercles, which has been hypothesized as an early sign of knee osteoarthritis. Additionally, we developed a model based on neural networks for identifying the feature in plain radiographs. Our results indicate promise in including tibial spiking as an early feature of knee osteoarthritis, and the feature is identifiable automatically. However, the work in the current thesis is limited and should be validated by future work.
Main Author
Format
Theses
Master thesis
Published
2022
Subjects
The permanent address of the publication
https://urn.fi/URN:NBN:fi:jyu-202212125538Use this for linking
Language
English