Machine Learning Detection and Characterization of Splenic Injuries on Abdominal Computed Tomography

被引:11
作者
Hamghalam, Mohammad [1 ,2 ,3 ]
Moreland, Robert [4 ,5 ]
Gomez, David [6 ,7 ,8 ,9 ]
Simpson, Amber [1 ,2 ]
Lin, Hui Ming [4 ]
Jandaghi, Ali Babaei [4 ,5 ]
Tafur, Monica [4 ,5 ]
Vlachou, Paraskevi A. [4 ,5 ]
Wu, Matthew [4 ,5 ]
Brassil, Michael [4 ,5 ]
Crivellaro, Priscila [4 ,5 ]
Mathur, Shobhit [4 ,5 ,7 ]
Hosseinpour, Shahob [4 ,5 ]
Colak, Errol [4 ,5 ,7 ]
机构
[1] Queens Univ, Sch Comp, Kingston, ON, Canada
[2] Queens Univ, Dept Biomed & Mol Sci, Kingston, ON, Canada
[3] Islamic Azad Univ, Dept Elect Engn, Qazvin Branch, Qazvin, Iran
[4] Unity Hlth Toronto, St Michaels Hosp, Dept Med Imaging, Toronto, ON, Canada
[5] Univ Toronto, Dept Med Imaging, Toronto, ON, Canada
[6] Unity Hlth Toronto, St Michaels Hosp, Div Gen Surg, Toronto, ON, Canada
[7] Unity Hlth Toronto, Li Ka Shing Knowledge Inst, Toronto, ON, Canada
[8] Univ Toronto, Temetry Fac Med, Dept Surg, Toronto, ON, Canada
[9] Univ Toronto, Inst Hlth Policy Management & Evaluat, Toronto, ON, Canada
来源
CANADIAN ASSOCIATION OF RADIOLOGISTS JOURNAL-JOURNAL DE L ASSOCIATION CANADIENNE DES RADIOLOGISTES | 2024年 / 75卷 / 03期
基金
加拿大自然科学与工程研究理事会;
关键词
splenic injury; machine learning; computed tomography; The American Association for the Surgery of Trauma; AAST; injury scoring scale; NONOPERATIVE MANAGEMENT; EMBOLIZATION; FAILURE;
D O I
10.1177/08465371231221052
中图分类号
R8 [特种医学]; R445 [影像诊断学];
学科分类号
1002 ; 100207 ; 1009 ;
摘要
Background: Multi-detector contrast-enhanced abdominal computed tomography (CT) allows for the accurate detection and classification of traumatic splenic injuries, leading to improved patient management. Their effective use requires rapid study interpretation, which can be a challenge on busy emergency radiology services. A machine learning system has the potential to automate the process, potentially leading to a faster clinical response. This study aimed to create such a system.Method: Using the American Association for the Surgery of Trauma (AAST), spleen injuries were classified into 3 classes: normal, low-grade (AAST grade I-III) injuries, and high-grade (AAST grade IV and V) injuries. Employing a 2-stage machine learning strategy, spleens were initially segmented from input CT images and subsequently underwent classification via a 3D dense convolutional neural network (DenseNet).Results: This single-centre retrospective study involved trauma protocol CT scans performed between January 1, 2005, and July 31, 2021, totaling 608 scans with splenic injuries and 608 without. Five board-certified fellowship-trained abdominal radiologists utilizing the AAST injury scoring scale established ground truth labels. The model achieved AUC values of 0.84, 0.69, and 0.90 for normal, low-grade injuries, and high-grade splenic injuries, respectively.Conclusions: Our findings demonstrate the feasibility of automating spleen injury detection using our method with potential applications in improving patient care through radiologist worklist prioritization and injury stratification. Future endeavours should concentrate on further enhancing and optimizing our approach and testing its use in a real-world clinical environment. Visual Abstract This is a visual representation of the abstract. Contexte: La tomodensitom & eacute;trie & agrave; multid & eacute;tecteurs avec injection de produit de contraste est une modalit & eacute; d'imagerie qui permet de d & eacute;tecter et de classifier les l & eacute;sions traumatiques de la rate de fa & ccedil;on satisfaisante, ce qui permet une meilleure prise en charge des patients. Cette modalit & eacute; d'imagerie exige une interpr & eacute;tation rapide des r & eacute;sultats pour & ecirc;tre utilis & eacute;e avec efficacit & eacute;, ce qui peut s'av & eacute;rer probl & eacute;matique dans le cadre d'un service de radiologie d'urgence aux journ & eacute;es bien charg & eacute;es. La mise au point d'un syst & egrave;me d'apprentissage pourrait mener & agrave; une automatisation de la proc & eacute;dure d'interpr & eacute;tation, et ainsi acc & eacute;l & eacute;rer la prise en charge clinique. La pr & eacute;sente & eacute;tude visait & agrave; concevoir un tel syst & egrave;me d'apprentissage.M & eacute;thode: Des l & eacute;sions de la rate ont & eacute;t & eacute; r & eacute;parties au sein de trois cat & eacute;gories au moyen du syst & egrave;me de classification des blessures traumatiques fourni par l'American Association for the Surgery of Trauma (AAST). Ces trois cat & eacute;gories & eacute;taient : organe normal, l & eacute;sion de bas grade (grades I, II et III de l'AAST) et l & eacute;sion de haut grade (grades IV et V de l'AAST). Notre strat & eacute;gie a & eacute;t & eacute; de recourir & agrave; deux syst & egrave;mes d'apprentissage : lors de la premi & egrave;re & eacute;tape, les images de rates obtenues par TDM ont & eacute;t & eacute; segment & eacute;es, puis, lors de la deuxi & egrave;me & eacute;tape, les l & eacute;sions ont & eacute;t & eacute; class & eacute;es & agrave; l'aide d'un r & eacute;seau neuronal convolutif 3D (DenseNet).R & eacute;sultats: Un total de 608 images de l & eacute;sions de la rate et 608 images de rates saines, obtenues au moyen d'examens de TDM dans le cadre d'un protocole d'& eacute;valuation des traumatismes entre le 1er janvier 2005 et le 31 juillet 2021, ont & eacute;t & eacute; incluses dans cette & eacute;tude r & eacute;trospective unicentrique. Cinq radiologistes d & eacute;tenteurs d'un certificat de sp & eacute;cialiste en radiologie abdominale ayant effectu & eacute; une formation de type fellowship se sont servis de l'& eacute;chelle d'& eacute;valuation des blessures de l'AAST pour attribuer un grade de r & eacute;f & eacute;rence & agrave; chaque image. Notre mod & egrave;le d'interpr & eacute;tation a obtenu un score AUC de 0,84 dans le cas des organes normaux, de 0,69 dans le cas des l & eacute;sions de bas grade et de 0,90 dans le cas des l & eacute;sions de haut grade.Conclusions: Les r & eacute;sultats indiquent qu'il est possible de d & eacute;tecter de fa & ccedil;on automatique des l & eacute;sions de la rate au moyen de notre m & eacute;thode. Celle-ci pr & eacute;sente des applications potentielles li & eacute;es & agrave; l'& eacute;tablissement des priorit & eacute;s des t & acirc;ches en radiologie et & agrave; la classification des l & eacute;sions, le tout en vue d'am & eacute;liorer les soins aux patients. De futures recherches devraient porter sur l'optimisation de notre m & eacute;thode et l'analyse de son utilisation dans un environnement clinique lors de situations r & eacute;elles.
引用
收藏
页码:534 / 541
页数:8
相关论文
共 18 条
[1]   Interrater Agreement of CT Grading of Blunt Splenic Injuries: Does the AAST Grading Need to Be Reimagined? [J].
Adams-McGavin, R. Chris ;
Tafur, Monica ;
Vlachou, Paraskevi A. ;
Wu, Matthew ;
Brassil, Michael ;
Crivellaro, Priscila ;
Lin, Hui-Ming ;
Gomez, David ;
Colak, Errol .
CANADIAN ASSOCIATION OF RADIOLOGISTS JOURNAL-JOURNAL DE L ASSOCIATION CANADIENNE DES RADIOLOGISTES, 2024, 75 (01) :171-177
[2]   Inter- and intrarater reliability in computed axial tomographic grading of splenic injury: Why so many grading scales? [J].
Barquist, ES ;
Pizano, LR ;
Feuer, W ;
Pappas, PA ;
McKenney, KA ;
LeBlang, SD ;
Henry, RP ;
Rivas, LA ;
Cohn, SM .
JOURNAL OF TRAUMA-INJURY INFECTION AND CRITICAL CARE, 2004, 56 (02) :334-338
[3]   Meta-analysis of predictive factors and outcomes for failure of non-operative management of blunt splenic trauma [J].
Bhangu, Aneel ;
Nepogodiev, Dmitri ;
Lal, Neeraj ;
Bowley, Douglas M. .
INJURY-INTERNATIONAL JOURNAL OF THE CARE OF THE INJURED, 2012, 43 (09) :1337-1346
[4]   Selective angiographic embolization of blunt splenic traumatic injuries in adults decreases failure rate of nonoperative management [J].
Bhullar, Indermeet S. ;
Frykberg, Eric R. ;
Siragusa, Daniel ;
Chesire, David ;
Paul, Julia ;
Tepas, Joseph J., III ;
Kerwin, Andrew J. .
JOURNAL OF TRAUMA AND ACUTE CARE SURGERY, 2012, 72 (05) :1127-1134
[5]   Optimizing Trauma Multidetector CT Protocol for Blunt Splenic Injury: Need for Arterial and Portal Venous Phase Scans [J].
Boscak, Alexis R. ;
Shanmuganathan, Kathirkamanathan ;
Mirvis, Stuart E. ;
Fleiter, Thorsten R. ;
Miller, Lisa A. ;
Sliker, Clint W. ;
Steenburg, Scott D. ;
Alexander, Melvin .
RADIOLOGY, 2013, 268 (01) :79-88
[6]  
Chahine AH., 2021, Acad Radiol, V28
[7]   Toward automated interpretable AAST grading for blunt splenic injury [J].
Chen, Haomin ;
Unberath, Mathias ;
Dreizin, David .
EMERGENCY RADIOLOGY, 2023, 30 (01) :41-50
[8]   The three-dimensional weakly supervised deep learning algorithm for traumatic splenic injury detection and sequential localization: an experimental study [J].
Cheng, Chi-Tung ;
Lin, Hou-Shian ;
Hsu, Chih-Po ;
Chen, Huan-Wu ;
Huang, Jen-Fu ;
Fu, Chih-Yuan ;
Hsieh, Chi-Hsun ;
Yeh, Chun-Nan ;
Chung, I-Fang ;
Liao, Chien-Hung .
INTERNATIONAL JOURNAL OF SURGERY, 2023, 109 (05) :1115-1124
[9]   CT grading scales for splenic injury: Why can't we agree? [J].
Clark, Rhys ;
Hird, Kathryn ;
Misur, Philip ;
Ramsay, Duncan ;
Mendelson, Richard .
JOURNAL OF MEDICAL IMAGING AND RADIATION ONCOLOGY, 2011, 55 (02) :163-169
[10]   The role of splenic angioembolization as an adjunct to nonoperative management of blunt splenic injuries: A systematic review and meta-analysis [J].
Crichton, James Charles Ian ;
Naidoo, Kamil ;
Yet, Barbaros ;
Brundage, Susan I. ;
Perkins, Zane .
JOURNAL OF TRAUMA AND ACUTE CARE SURGERY, 2017, 83 (05) :934-943