Requirements for artificial intelligence used for the diagnosis of the COVID-19 from chest X-rays
Authors
Date
2022Access restrictions
The author has not given permission to make the work publicly available electronically. Therefore the material can be read only at the archival workstation at Jyväskylä University Library (https://kirjasto.jyu.fi/collections/archival-workstation).
Copyright
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Viime aikoina on julkaistu useita tekoälymalleja COVID-19 diagnosoimiseen keuhkoröntgenkuvista. Valitettavasti useiden arvioiden perusteella niissä on ongelmia jotka tekevät ne käyttökelvottomiksi kliinisessä työssä. Tässä työssä on kerätty eri lähteistä tietoa siitä mitä tekoälymallilta vaaditaan jotta sitä voi käyttää diagnosoimiseen.
Tärkeimmät vaatimukset ovat tuloksen selitettävyys, käytetyn datan vääristymien korjaaminen (esimerkiksi ikäjakaumat voivat olla hyvin erilaiset eri lähteissä), tarkka dokumentaatio prosessista ja perinpohjainen tilastollinen analyysi mallin toimivuudesta.
Työn heikkoutena voidaan pitää käytettyjen metodien rajoittumista aikaisempien julkaisujen tutkimiseen ja sitä että käytössä ei ollut lääketieteellistä asiantuntemusta. Mahdolliset lisävaatimukset eivät kuitenkaan kumoa tässä työssä löydettyjä vaatimuksia. There has been multiple publications with artificial intelligence (AI) models for COVID-19 diagnosis using chest X-ray images. Unfortunately according to multiple reviews the suggested models have issues making them unusable for clinical application. This work has collected the requirements for a diagnostic AI model using various sources. The most important requirements are: explainability, bias corrections (for example age distribution can have significant differences between datasets), precise documentation and thorough statistical analysis of the performance of the proposed model.
The main weaknesses in this work are the choice of methods used, as they are limited to study of previous publications, and lack of available clinical expertise. This is not critical issue as possible additional requirements will not be in conflict with the requirements found in this work.
Keywords
Metadata
Show full item recordCollections
- Pro gradu -tutkielmat [29605]
Related items
Showing items with similar title or keywords.
-
Requirement analysis for an artificial intelligence model for the diagnosis of the COVID-19 from chest X-ray data
Kalliokoski, Tuomo (IEEE, 2021)There are multiple papers published about different AI models for the COVID-19 diagnosis with promising results. Unfortunately according to the reviews many of the papers do not reach the level of sophistication needed for ... -
Robustness, Stability, and Fidelity of Explanations for a Deep Skin Cancer Classification Model
Saarela, Mirka; Geogieva, Lilia (MDPI AG, 2022)Skin cancer is one of the most prevalent of all cancers. Because of its being widespread and externally observable, there is a potential that machine learning models integrated into artificial intelligence systems will ... -
Strategic cyber threat intelligence : Building the situational picture with emerging technologies
Voutilainen, Janne; Kari, Martti (Academic Conferences International, 2020)In 2019, e-criminals adopted new tactics to demand enormous ransoms from large organizations by using ransomware, a phenomenon known as “big game hunting.” Big game hunting is an excellent example of a sophisticated and ... -
Toward a Regulatory-Compliant Lifecycle for Artificial-Intelligence-Based Medical Devices in the European Union : Industry Perspectives
Granlund, Tuomas; Stirbu, Vlad; Mikkonen, Tommi (IEEE, 2024)Despite the immense potential of artificial intelligence (AI)-powered medical devices to revolutionize health care, concerns regarding their safety in life-critical applications remain. This article proposes extending the ... -
On Attacking Future 5G Networks with Adversarial Examples : Survey
Zolotukhin, Mikhail; Zhang, Di; Hämäläinen, Timo; Miraghaei, Parsa (MDPI AG, 2023)The introduction of 5G technology along with the exponential growth in connected devices is expected to cause a challenge for the efficient and reliable network resource allocation. Network providers are now required to ...