Understanding metric-related pitfalls in image analysis validation

被引：35

作者：

Reinke, Annika ^{[1
,2
,3
]}

Tizabi, Minu D. ^{[1
,4
]}

Baumgartner, Michael ^{[5
]}

Eisenmann, Matthias ^{[1
]}

Heckmann-Noetzel, Doreen ^{[1
,4
]}

Kavur, A. Emre ^{[1
,5
,6
]}

Raedsch, Tim ^{[1
,2
]}

Sudre, Carole H. ^{[7
,8
,9
]}

Acion, Laura ^{[10
]}

Antonelli, Michela ^{[9
,11
]}

Arbel, Tal ^{[12
,13
]}

Bakas, Spyridon ^{[14
,15
]}

Benis, Arriel ^{[16
,17
]}

Buettner, Florian ^{[18
,19
,20
,21
,22
]}

Cardoso, M. Jorge ^{[9
]}

Cheplygina, Veronika ^{[23
]}

Chen, Jianxu ^{[24
]}

Christodoulou, Evangelia ^{[1
]}

Cimini, Beth A. ^{[25
]}

Farahani, Keyvan ^{[26
]}

Ferrer, Luciana ^{[27
]}

Galdran, Adrian ^{[28
,29
]}

van Ginneken, Bram ^{[30
,31
]}

Glocker, Ben ^{[32
]}

Godau, Patrick ^{[1
,3
,4
]}

Hashimoto, Daniel A. ^{[33
,34
]}

Hoffman, Michael M. ^{[35
,36
,37
,38
]}

Huisman, Merel ^{[39
]}

Isensee, Fabian ^{[5
,6
]}

Jannin, Pierre ^{[40
,41
]}

Kahn, Charles E. ^{[42
,43
]}

Kainmueller, Dagmar ^{[44
,45
,46
]}

Kainz, Bernhard ^{[47
,48
]}

Karargyris, Alexandros ^{[49
]}

Kleesiek, Jens ^{[50
]}

Kofler, Florian ^{[51
]}

Kooi, Thijs ^{[52
]}

Kopp-Schneider, Annette ^{[53
]}

Kozubek, Michal ^{[54
,55
]}

Kreshuk, Anna ^{[56
]}

Kurc, Tahsin ^{[57
]}

Landman, Bennett A. ^{[58
]}

Litjens, Geert ^{[59
]}

Madani, Amin ^{[60
]}

Maier-Hein, Klaus ^{[5
,61
]}

Martel, Anne L. ^{[36
,62
]}

Meijering, Erik ^{[63
]}

Menze, Bjoern ^{[64
]}

Moons, Karel G. M. ^{[65
]}

Mueller, Henning ^{[66
,67
]}

机构：

[1] German Canc Res Ctr DKFZ Heidelberg, Div Intelligent Med Syst, Heidelberg, Germany

[2] German Canc Res Ctr DKFZ Heidelberg, HI Helmholtz Imaging, Heidelberg, Germany

[3] Heidelberg Univ, Fac Math & Comp Sci, Heidelberg, Germany

[4] NCT Heidelberg, Natl Ctr Tumor Dis NCT, Heidelberg, Germany

[5] German Canc Res Ctr DKFZ Heidelberg, Div Med Image Comp, Heidelberg, Germany

[6] German Canc Res Ctr DKFZ Heidelberg, HI Appl Comp Vis Lab, Heidelberg, Germany

[7] UCL, MRC Unit Lifelong Hlth & Ageing, London, England

[8] UCL, Dept Comp Sci, Ctr Med Image Comp, London, England

[9] Kings Coll London, Sch Biomed Engn & Imaging Sci, London, England

[10] Univ Buenos Aires, Inst Calculo, CONICET, Buenos Aires, DF, Argentina

[11] UCL, Ctr Med Image Comp, London, England

[12] McGill Univ, Ctr Intelligent Machines, Montreal, PQ, Canada

[13] McGill Univ, MILA Quebec Artificial Intelligence Inst, Montreal, PQ, Canada

[14] Indiana Univ Sch Med, Dept Pathol & Lab Med, Div Computat Pathol, Indianapolis, IN 46202 USA

[15] Univ Penn, Ctr Biomed Image Comp & Analyt CBICA, Philadelphia, PA 19104 USA

[16] Holon Inst Technol, Dept Digital Med Technol, Holon, Israel

[17] European Federat Med Informat, Le Mt Sur Lausanne, Switzerland

[18] German Canc Consortium DKTK, Partner Site Frankfurt Mainz, Frankfurt, Germany

[19] German Canc Res Ctr DKFZ Heidelberg, Heidelberg, Germany

[20] Goethe Univ Frankfurt, Dept Med, Frankfurt, Germany

[21] Goethe Univ Frankfurt, Dept Informat, Frankfurt, Germany

[22] Frankfurt Canc Insititute, Frankfurt, Germany

[23] IT Univ Copenhagen, Dept Comp Sci, Copenhagen, Denmark

[24] Leibniz Inst Analyt Wissensch ISAS eV, Dortmund, Germany

[25] Broad Inst MIT & Harvard, Imaging Platform, Cambridge, MA 02142 USA

[26] NCI, Ctr Biomed Informat & Informat Technol, Bethesda, MD 20892 USA

[27] UBA, CONICET, Inst Invest Ciencias Computac ICC, Buenos Aires, DF, Argentina

[28] Univ Pompeu Fabra, Barcelona, Spain

[29] Univ Adelaide, Adelaide, SA, Australia

[30] Fraunhofer MEVIS, Bremen, Germany

[31] Radboud Univ Nijmegen, Med Ctr, Radboud Inst Hlth Sci, Nijmegen, Netherlands

[32] Imperial Coll London, Dept Comp, South Kensington Campus, London, England

[33] Perelman Sch Med, Dept Surg, Philadelphia, PA USA

[34] Univ Penn, Sch Engn & Appl Sci, Gen Robot Automat Sensing & Percept Lab, Philadelphia, PA 19104 USA

[35] Univ Hlth Network, Princess Margaret Canc Ctr, Toronto, ON, Canada

[36] Univ Toronto, Dept Med Biophys, Toronto, ON, Canada

[37] Univ Toronto, Dept Comp Sci, Toronto, ON, Canada

[38] Vector Inst Artificial Intelligence, Toronto, ON, Canada

[39] Radboud Univ Nijmegen, Med Ctr, Dept Radiol & Nucl Med, Nijmegen, Netherlands

[40] Univ Rennes 1, Lab Traitement Signal & Image, UMR S 1099, Rennes, France

[41] INSERM, Paris, France

[42] Univ Penn, Dept Radiol, Philadelphia, PA 19104 USA

[43] Univ Penn, Inst Biomed Informat, Philadelphia, PA 19104 USA

[44] Max Delbruck Ctr Mol Med, Helmholtz Assoc MDC, Biomed Image Anal, Berlin, Germany

[45] HI Helmholtz Imaging, Berlin, Germany

[46] Univ Potsdam, Digital Engn Fac, Potsdam, Germany

[47] Imperial Coll London, Fac Engn, Dept Comp, London, England

[48] Friedrich Alexander Univ, Dept AIBE, Erlangen, Germany

[49] IHU Strasbourg, Strasbourg, France

[50] Univ Med Essen, Inst AI Med IKIM, Translat Image Guided Oncol TIO, Essen, Germany

来源：

NATURE METHODS | 2024年 / 21卷 / 02期

基金：

芬兰科学院; 瑞士国家科学基金会; 加拿大自然科学与工程研究理事会; 英国工程与自然科学研究理事会; “创新英国”项目; 美国国家卫生研究院; 欧洲研究理事会; 英国惠康基金; 荷兰研究理事会;

关键词：

SEGMENTATION;

D O I：

10.1038/s41592-023-02150-0

中图分类号：

Q5 [生物化学];

学科分类号：

071010 ; 081704 ;

摘要：

Validation metrics are key for tracking scientific progress and bridging the current chasm between artificial intelligence research and its translation into practice. However, increasing evidence shows that, particularly in image analysis, metrics are often chosen inadequately. Although taking into account the individual strengths, weaknesses and limitations of validation metrics is a critical prerequisite to making educated choices, the relevant knowledge is currently scattered and poorly accessible to individual researchers. Based on a multistage Delphi process conducted by a multidisciplinary expert consortium as well as extensive community feedback, the present work provides a reliable and comprehensive common point of access to information on pitfalls related to validation metrics in image analysis. Although focused on biomedical image analysis, the addressed pitfalls generalize across application domains and are categorized according to a newly created, domain-agnostic taxonomy. The work serves to enhance global comprehension of a key topic in image analysis validation. This Perspective presents a reliable and comprehensive source of information on pitfalls related to validation metrics in image analysis, with an emphasis on biomedical imaging.

引用

页码：182 / 194

页数：20

共 38 条

[1] The Liver Tumor Segmentation Benchmark (LiTS)
Bilic, Patrick
Christ, Patrick
Li, Hongwei Bran
Vorontsov, Eugene
Ben-Cohen, Avi
Kaissis, Georgios
Szeskin, Adi
Jacobs, Colin
Mamani, Gabriel Efrain Humpire
Chartrand, Gabriel
Lohoefer, Fabian
Holch, Julian Walter
Sommer, Wieland
Hofmann, Felix
Hostettler, Alexandre
Lev-Cohain, Naama
Drozdzal, Michal
Amitai, Michal Marianne
Vivanti, Refael
Sosna, Jacob
Ezhov, Ivan
Sekuboyina, Anjany
Navarro, Fernando
Kofler, Florian
Paetzold, Johannes C.
Shit, Suprosanna
Hu, Xiaobin
Lipkova, Jana
Rempfler, Markus
Piraud, Marie
Kirschke, Jan
Wiestler, Benedikt
Zhang, Zhiheng
Huelsemeyer, Christian
Beetz, Marcel
Ettlinger, Florian
Antonelli, Michela
Bae, Woong
Bellver, Miriam
Bi, Lei
Chen, Hao
Chlebus, Grzegorz
Dam, Erik B.
Dou, Qi
Fu, Chi-Wing
Georgescu, Bogdan
Giro-I-Nieto, Xavier
Gruen, Felix
Han, Xu
Heng, Pheng-Ann
[J]. MEDICAL IMAGE ANALYSIS, 2023, 84
[2] Brown B., 1968, P-3925
[3] Effects of the trinucleotide preceding the self-cleavage site on eggplant latent viroid hammerheads:: differences in co- and post-transcriptional self-cleavage may explain the lack of trinucleotide AUC in most natural hammerheads
Carbonell, Alberto
De la Pena, Marcos
Flores, Ricardo
Gago, Selma
[J]. NUCLEIC ACIDS RESEARCH, 2006, 34 (19) : 5613 - 5622
[4] Chen J., 2020, bioRxiv, DOI DOI 10.1101/491035
[5] The Matthews correlation coefficient (MCC) is more reliable than balanced accuracy, bookmaker informedness, and markedness in two-class confusion matrix evaluation
Chicco, Davide
Totsch, Niklas
Jurman, Giuseppe
[J]. BIODATA MINING, 2021, 14 (01) : 1 - 22
[6] The advantages of the Matthews correlation coefficient (MCC) over F1 score and accuracy in binary classification evaluation
Chicco, Davide
Jurman, Giuseppe
[J]. BMC GENOMICS, 2020, 21 (01)
[7] Cordts M., 2015, CVPR WORKSH FUT DAT, V2
[8] Correia Paulo, 2006, EURASIP Journal on Advances in Signal Processing, V2006, P1
[9] Nonceliac Gluten Sensitivity: Sense or Sensibility?
Di Sabatino, Antonio
Corazza, Gino Roberto
[J]. ANNALS OF INTERNAL MEDICINE, 2012, 156 (04) : 309 - U101
[10] The Pascal Visual Object Classes (VOC) Challenge
Everingham, Mark
Van Gool, Luc
Williams, Christopher K. I.
Winn, John
Zisserman, Andrew
[J]. INTERNATIONAL JOURNAL OF COMPUTER VISION, 2010, 88 (02) : 303 - 338

← 1 2 3 4 →