Deep learning-based decision forest for hereditary clear cell renal cell carcinoma segmentation on MRI

被引:4
作者
Lay, Nathan [1 ]
Anari, Pouria Yazdian [2 ]
Chaurasia, Aditi [2 ]
Firouzabadi, Fatemeh Dehghani [2 ]
Harmon, Stephanie [1 ]
Turkbey, Evrim [2 ]
Gautam, Rabindra [3 ]
Samimi, Safa [2 ]
Merino, Maria J. J. [4 ]
Ball, Mark W. W. [3 ]
Linehan, William Marston [3 ]
Turkbey, Baris [1 ]
Malayeri, Ashkan A. A. [2 ]
机构
[1] NCI, Artificial Intelligence Resource, Mol Imaging Branch, Bethesda, MD 20892 USA
[2] NIH, Radiol & Imaging Sci, Clin Ctr, Bethesda, MD USA
[3] NCI, Urol Oncol Branch, Ctr Canc Res, Bethesda, MD USA
[4] NCI, Lab Pathol, Ctr Canc Res, Bethesda, MD USA
关键词
decision forest; deep learning; MRI; segmentation; von Hippel-Lindau;
D O I
10.1002/mp.16303
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
Backgroundvon Hippel-Lindau syndrome (VHL) is an autosomal dominant hereditary syndrome with an increased predisposition of developing numerous cysts and tumors, almost exclusively clear cell renal cell carcinoma (ccRCC). Considering the lifelong surveillance in such patients to monitor the disease, patients with VHL are preferentially imaged using MRI to eliminate radiation exposure. PurposeSegmentation of kidney and tumor structures on MRI in VHL patients is useful in lesion characterization (e.g., cyst vs. tumor), volumetric lesion analysis, and tumor growth prediction. However, automated tasks such as ccRCC segmentation on MRI is sparsely studied. We develop segmentation methodology for ccRCC on T1 weighted precontrast, corticomedullary, nephrogenic, and excretory contrast phase MRI. MethodsWe applied a new neural network approache using a novel differentiable decision forest, called hinge forest (HF), to segment kidney parenchyma, cyst, and ccRCC tumors in 117 images from 115 patients. This data set represented an unprecedented 504 ccRCCs with 1171 cystic lesions obtained at five different MRI scanners. The HF architecture was compared with U-Net on 10 randomized splits with 75% used for training and 25% used for testing. Both methods were trained with Adam using default parameters (alpha=0.001,beta 1=0.9,beta 2=0.999$\alpha = 0.001,\ \beta _1 = 0.9,\ \beta _2 = 0.999$) over 1000 epochs. We further demonstrated some interpretability of our HF method by exploiting decision tree structure. ResultsThe HF achieved an average kidney, cyst, and tumor Dice similarity coefficient (DSC) of 0.75 +/- 0.03, 0.44 +/- 0.05, 0.53 +/- 0.04, respectively, while U-Net achieved an average kidney, cyst, and tumor DSC of 0.78 +/- 0.02, 0.41 +/- 0.04, 0.46 +/- 0.05, respectively. The HF significantly outperformed U-Net on tumors while U-Net significantly outperformed HF when segmenting kidney parenchymas (alpha<0.01$\alpha < 0.01$). ConclusionsFor the task of ccRCC segmentation, the HF can offer better segmentation performance compared to the traditional U-Net architecture. The leaf maps can glean hints about deep learning features that might prove to be useful in other automated tasks such as tumor characterization.
引用
收藏
页码:5020 / 5029
页数:10
相关论文
共 26 条
[1]  
Agarwal A., 2021, THESIS U GUELPH
[2]   Association of genomic subtypes of lower-grade gliomas with shape features automatically extracted by a deep learning algorithm [J].
Buda, Mateusz ;
Saha, Ashirbani ;
Mazurowski, Maciej A. .
COMPUTERS IN BIOLOGY AND MEDICINE, 2019, 109 :218-225
[3]  
Demsar J, 2006, J MACH LEARN RES, V7, P1
[4]  
Frosst N, 2017, Distilling a Neural Network Into a Soft Decision Tree
[5]   THE RICIAN DISTRIBUTION OF NOISY MRI DATA [J].
GUDBJARTSSON, H ;
PATZ, S .
MAGNETIC RESONANCE IN MEDICINE, 1995, 34 (06) :910-914
[6]  
Hastie T., 2009, The Elements of Statistical Learning, V2
[7]  
Hazimeh H, 2020, PR MACH LEARN RES, V119
[8]   Comparison and Evaluation of Methods for Liver Segmentation From CT Datasets [J].
Heimann, Tobias ;
van Ginneken, Bram ;
Styner, Martin A. ;
Arzhaeva, Yulia ;
Aurich, Volker ;
Bauer, Christian ;
Beck, Andreas ;
Becker, Christoph ;
Beichel, Reinhard ;
Bekes, Gyoergy ;
Bello, Fernando ;
Binnig, Gerd ;
Bischof, Horst ;
Bornik, Alexander ;
Cashman, Peter M. M. ;
Chi, Ying ;
Cordova, Andres ;
Dawant, Benoit M. ;
Fidrich, Marta ;
Furst, Jacob D. ;
Furukawa, Daisuke ;
Grenacher, Lars ;
Hornegger, Joachim ;
Kainmueller, Dagmar ;
Kitney, Richard I. ;
Kobatake, Hidefumi ;
Lamecker, Hans ;
Lange, Thomas ;
Lee, Jeongjin ;
Lennon, Brian ;
Li, Rui ;
Li, Senhu ;
Meinzer, Hans-Peter ;
Nemeth, Gabor ;
Raicu, Daniela S. ;
Rau, Anne-Mareike ;
van Rikxoort, Eva M. ;
Rousson, Mikael ;
Rusko, Laszlo ;
Saddi, Kinda A. ;
Schmidt, Guenter ;
Seghers, Dieter ;
Shimizu, Akinobu ;
Slagmolen, Pieter ;
Sorantin, Erich ;
Soza, Grzegorz ;
Susomboon, Ruchaneewan ;
Waite, Jonathan M. ;
Wimmer, Andreas ;
Wolf, Ivo .
IEEE TRANSACTIONS ON MEDICAL IMAGING, 2009, 28 (08) :1251-1265
[9]  
Heller N, 2019, The KiTS19 challenge data: 300 kidney tumor cases with clinical context, CT semantic segmentations, and surgical outcomes
[10]   The state of the art in kidney and kidney tumor segmentation in contrast-enhanced CT imaging: Results of the KiTS19 challenge [J].
Heller, Nicholas ;
Isensee, Fabian ;
Maier-Hein, Klaus H. ;
Hou, Xiaoshuai ;
Xie, Chunmei ;
Li, Fengyi ;
Nan, Yang ;
Mu, Guangrui ;
Lin, Zhiyong ;
Han, Miofei ;
Yao, Guang ;
Gao, Yaozong ;
Zhang, Yao ;
Wang, Yixin ;
Hou, Feng ;
Yang, Jiawei ;
Xiong, Guangwei ;
Tian, Jiang ;
Zhong, Cheng ;
Ma, Jun ;
Rickman, Jack ;
Dean, Joshua ;
Stai, Bethany ;
Tejpaul, Resha ;
Oestreich, Makinna ;
Blake, Paul ;
Kaluzniak, Heather ;
Raza, Shaneabbas ;
Rosenberg, Joel ;
Moore, Keenan ;
Walczak, Edward ;
Rengel, Zachary ;
Edgerton, Zach ;
Vasdev, Ranveer ;
Peterson, Matthew ;
McSweeney, Sean ;
Peterson, Sarah ;
Kalapara, Arveen ;
Sathianathen, Niranjan ;
Papanikolopoulos, Nikolaos ;
Weight, Christopher .
MEDICAL IMAGE ANALYSIS, 2021, 67