A review of evaluation approaches for explainable AI with applications in cardiology

被引：10

作者：

Salih, Ahmed M. ^{[1
,2
,3
]}

Galazzo, Ilaria Boscolo ^{[4
]}

Gkontra, Polyxeni ^{[5
]}

Rauseo, Elisa ^{[1
]}

Lee, Aaron Mark ^{[1
]}

Lekadir, Karim ^{[5
,6
]}

Radeva, Petia ^{[7
]}

Petersen, Steffen E. ^{[1
,8
,9
,10
]}

Menegaz, Gloria ^{[4
]}

机构：

[1] Queen Mary Univ London, William Harvey Res Inst, NIHR Barts Biomed Res Ctr, Charterhouse Sq, London EC1M 6BQ, England

[2] Univ Leicester, Dept Populat Hlth Sci, Univ Rd, Leicester LE1 7RH, England

[3] Univ Zakho, Dept Comp Sci, Duhok Rd, Zakho, Kurdistan, Iraq

[4] Univ Verona, Dept Engn Innovat Med, S Francesco 22, I-37129 Verona, Italy

[5] Univ Barcelona, Fac Math & Comp Sci, Artificial Intelligence Med Lab BCN AIM, Gran Via Corts Catalanes 585, Barcelona 08007, Barcelona, Spain

[6] Inst Catalana Recerca & Estudis Avancats ICREA, Passeig Lluis Co 23, Barcelona 08010, Spain

[7] Univ Barcelona, Dept Matemat & Informat, Gran Via Corts Catalanes 585, Barcelona 08007, Spain

[8] St Bartholomews Hosp, Barts Heart Ctr, Barts Hlth NHS Trust, London, England

[9] Hlth Data Res, London, England

[10] Alan Turing Inst, London, England

来源：

ARTIFICIAL INTELLIGENCE REVIEW | 2024年 / 57卷 / 09期

关键词：

Cardiac; AI; XAI; Evaluation; ARTIFICIAL-INTELLIGENCE; PULMONARY-HYPERTENSION; ATRIAL-FIBRILLATION; RISK PREDICTION; LEARNING-MODEL; CARDIAC EVENTS; HEART-FAILURE; MACHINE; ELECTROCARDIOGRAM; MORTALITY;

D O I：

10.1007/s10462-024-10852-w

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

Explainable artificial intelligence (XAI) elucidates the decision-making process of complex AI models and is important in building trust in model predictions. XAI explanations themselves require evaluation as to accuracy and reasonableness and in the context of use of the underlying AI model. This review details the evaluation of XAI in cardiac AI applications and has found that, of the studies examined, 37% evaluated XAI quality using literature results, 11% used clinicians as domain-experts, 11% used proxies or statistical analysis, with the remaining 43% not assessing the XAI used at all. We aim to inspire additional studies within healthcare, urging researchers not only to apply XAI methods but to systematically assess the resulting explanations, as a step towards developing trustworthy and safe models.

引用

页数：44

共 263 条

[1] Explaining individual predictions when features are dependent: More accurate approximations to Shapley values [J].