Dual segmentation models for poorly and well-differentiated hepatocellular carcinoma using two-step transfer deep learning on dynamic contrast-enhanced CT images

doi:10.21203/rs.3.rs-1749800/v1

Download PDF

Research Article

Dual segmentation models for poorly and well-differentiated hepatocellular carcinoma using two-step transfer deep learning on dynamic contrast-enhanced CT images

https://doi.org/10.21203/rs.3.rs-1749800/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Aim: The aim of this study was to develop dual segmentation models for poorly and well-differentiated hepatocellular carcinoma (HCC), using two-step transfer learning (TSTL) based on dynamic contrast-enhanced (DCE) computed tomography (CT) images.

Methods: From 2013 to 2019, DCE CT images of 128 patients with 80 poorly differentiated and 48 well-differentiated HCCs were selected at our hospital. In the first transfer learning (TL) step, a pre-trained segmentation model with 192 CT images of lung cancer patients was retrained as a poorly differentiated HCC model. In the second TL step, a well-differentiated HCC model was built from a poorly differentiated HCC model. The average 3D Dice’s similarity coefficient (3D-DSC) and 95th-percentile of the Hausdorff distance (95% HD) were employed to evaluate the segmentation accuracy, based on a nested 4-fold cross-validation test. The DSC denotes the degree of regional similarity between the HCC reference regions and the regions estimated using the proposed models. The 95% HD is defined as the 95th-percentile of the maximum measures of how far two subsets of a metric space are from each other.

Results: The average 3D-DSC and 95% HD were 0.849 ± 0.078 and 1.98 ± 0.71 mm, respectively, for poorly differentiated HCC regions, and 0.811 ± 0.089 and 2.01 ± 0.84 mm, respectively, for well-differentiated HCC regions. The average 3D-DSC for both regions was 1.2 times superior to that calculated without the TSTL.

Conclusion: The proposed model using TSTL from the lung cancer dataset showed the potential to segment poorly and well-differentiated HCC regions on DCE CT images.

deep learning

hepatocellular carcinoma

dual segmentation

poorly differentiated

well-differentiated

transfer learning

Hepatocellular carcinoma (HCC) was the sixth most commonly diagnosed cancer in the world in 2020, with an estimated 815,000 new cases and 747,000 deaths annually [1]. Approximately 90% of primary liver cancers are HCC, which is the fourth leading cause of cancer-related deaths worldwide [2–4]. According to the Clinical Practice Guidelines for Hepatocellular Carcinoma 2017 [5], HCC treatment strategies are decided based on the Child-Pugh score, metastasis, and portal invasion status, number of nodules, and the size of the single largest dimension of enhancing viable tumor on arterial phase dynamic contrast-enhanced (DCE) computed tomography (CT) images. In particular, radiofrequency ablation (RFA) and transcatheter arterial chemoembolization (TACE) have been chosen depending on the largest tumor diameter (threshold value: 3 cm). Therefore, the accurate segmentation of HCC regions is substantially necessary for determination of treatment strategies. However, there are two types of HCCs, well and poorly differentiated HCC, which have different imaging features at their tumor peripheral regions in DCE CT images. Moreover, contrast enhancement patterns differ in the three phases of the arterial phase, portal vein phase, and equilibrium phase depending on the differentiation of the HCC [6]. That is because the arterial blood flows from angiogenesis could increase in the peripheral regions of poorly differentiated HCC [7]. Thus, separate segmentations of the poorly and well-differentiated HCC regions on DCE CT images could improve the accuracy. The degree of the differentiation could affect the treatment strategy and management of treatment progress [5]. Recurrence and metastasis are more likely to occur in poorly differentiated HCC with high pathologic malignancy grade than well-differentiated HCC with low pathologic malignancy grade [8]. In particular, the patients with poorly differentiated HCCs larger than 2 cm in diameter may have early recurrence within 2 years after their surgery [9]. These backgrounds also encourage us to investigate separate segmentation approaches for poorly and well-differentiated HCCs.

However, a study by Kim et al. [10] suggested that there is interobserver variability in the measurement of the maximum HCC tumor diameter (two observers' Kappa coefficient:0.56), which could lead to poor reliability in decision-making on treatment strategies. Therefore, automated segmentation of HCC regions is essential for accurately and reliably measuring the longest tumor diameters. Past studies have developed deep learning (DL) models for diagnosing the differentiation of HCCs and automated segmentation of HCC regions [11–15]. Yasaka et al [11] indicated a median accuracy of 0.84 for differential diagnosis of hepatic masses relative to DCE CT image data. Furthermore, the average DSCs for the segmentation accuracy of HCC in CT images obtained in these studies ranged from 0.761 to 0.825. Yuan et al. [12] studied a hierarchical framework based on deep fully convolutional deconvolutional neural networks to segment HCC regions. Li et al. [13] proposed a novel hybrid densely connected UNet (H-Dense Unet), which consists of a 2D dense Unet and a 3D counterpart for hierarchically aggregating volumetric contexts. Qayyum et al. [14] proposed a hybrid 3D residual network (RN) with a squeeze-and-excitation (SE) block for volumetric segmentation of the kidney, liver, and their associated tumors, such as HCC. Alirr [15] proposed a method for liver and liver tumor segmentation using a fully convolutional neural network (FCN) with region-based level set functions. In these past studies [12–15], the Liver Tumor Segmentation (LiTS) challenge dataset [16] was used, which was not categorized into poorly differentiated and well-differentiated HCCs.

In this study, we hypothesized that the separate segmentations of the poorly and well-differentiated HCC regions on DCE CT images could improve their accuracy, because the DCE CT images describe different features of poorly and well-differentiated HCC in the arterial, portal, and equilibrium phases, from a diagnostic point of view. Time-variant information from DCE CT images could be useful for prediction of microvascular invasion of HCC, especially peripheral regions [17]. This information could increase the segmentation accuracy of each HCC. However, previous studies [12–15] have not explored segmentation models for poorly and well-differentiated HCC regions separately based on the three phases. Since our aim was to separately segment the poorly and well-differentiated HCC regions using DCE CT data from the arterial, portal, and equilibrium phases, we did not employ the one phase LiTS challenge dataset.

This study investigated DL models for the dual segmentation of poorly and well-differentiated HCC regions by applying the models to HCC CT datasets in an institution. To mitigate the overfitting problem and to make the best use of clinical resources for lung cancer and HCC cases, we employed two-step transfer learning (TSTL). We also hypothesized that a DL model for lung cancer would be applicable to segmentation of poorly differentiated HCC, which could be retrained for segmenting well-differentiated HCC, because of their similarity. We applied the DL model for lung cancer segmentation developed by Cui et al. [18] Hence, the aim of this study was to develop dual segmentation models for poorly and well-differentiated HCC regions on DCE CT images using TSTL based on Dense V-net. From the clinical point of view, the longest diameters obtained from predicted HCC regions were compared with those from reference HCC regions.

The protocol of this retrospective study was approved by the institutional review boards of the Saga University Hospital (approval number: 2017-11-JINSOKU-07) and Kyushu University Hospital (2021 − 201). Figure 1 shows the overall workflow of this study for segmenting HCC regions using TSTL. DCE CT images and annotations for poorly and well-differentiated HCCs at three phases (arterial, portal, equilibrium phases) were inputted to the TSTL models after a preprocessing. After generating a pre-trained segmentation model using CT images of lung cancer patients, poorly and well-differentiated HCC regions were sequentially and separately segmented using TSTL. The average 3D Dice’s similarity coefficient (3D-DSC) and the 95th-percentile of the Hausdorff distance (95% HD) were employed for evaluating the segmentation accuracy based on a nested 4-fold cross-validation test [19–21]. Additionally, Spearman’s rank correlation coefficients and errors between predicted HCC regions and annotations were assessed with respect to the longest HCC diameters.

Clinical cases

Eighty patients with poorly differentiated HCC and 48 patients with well-differentiated HCC were selected from patients treated in our institutions from 2013 to 2019. The patients underwent resection, RFA, TACE, and molecularly targeted drug therapy [5]. Table 1 summarizes the clinical cases of poorly differentiated HCC and well-differentiated HCC enrolled in this study.

Table 1

Summary of clinical cases for poorly differentiated HCC and welldifferentiated HCC in this study.
	Poorly differentiated HCC	Well-differentiated HCC
Total number of cases	80	48
Age (years), min-max(median)	49–91 (68.5)	57–86 (70)
Sex
Male	45	30
Female	35	18
Treatment methods
RFA	23	12
Resection	30	19
TACE	25	10
Molecularly targeted drug	2	5
Follow-up	0	2
Average longest HCC diameter of annotations (mm)	20.9	19.2

HCC patients were scanned at an arterial phase (40 s after injection), a portal phase (65 s after injection), and an equilibrium phase (240 s after injection) on a multi-slice CT scanner (Aquilion ONE ViSION, 320 rows, Canon Medical Systems, Otawara, Japan; LightSpeed VCT VISION, 64 rows; GE Healthcare, Chicago, IL; SOMATOM Definition force, 256 rows, Siemens Medical Solutions, Erlangen, Germany). The DCE CT images were acquired with a tube voltage of 120 kV, a gantry rotation time of 0.5 s, matrix size from 512 × 512 × 150 to 512 × 512 × 210 pixels (hereinafter expressed as 512 × 512 × 150–210 pixels); pixel sizes of 0.628, 0.680, and 0.750 mm; and slice thicknesses of 0.72, 0.80, 1.00 mm, respectively. In matrix sizes of CT images, the first two values represent the number of in-plane (axial plane) pixels, and the third value is the number of slices (inferior‒superior direction). The image noise level (standard deviation [SD]) was kept within 6.0 HU on the CT scanner, by using an automatic exposure control technique. CT images were reconstructed with two consecutive algorithms: applying a reconstruction function for a standard abdominal region with a low-frequency enhancement function, followed by using an iterative reconstruction algorithm.

The HCC region contours were delineated as annotations (reference contours) to be VOI on the arterial phase CT images in poorly differentiated HCC and on the equilibrium phase in well-differentiated HCC, based on consensus between an abdominal radiologist (JN) and a medical physicist (NN), using open-source software: 3D Slicer (version 4.10.2) [22]. The annotations were manually registered with HCC regions on the other two phases by translating the annotations, because misregistration of DCE images at different phases occurred between the annotations and HCC regions due to respiratory motions. CT images for well-differentiated HCC regions were inverted with respect to the CT values to obtain higher CT values in the HCC regions than those in the background.

Preprocessing of HCC region datasets for CT images

Anisotropic CT and annotation binary images were transformed into isotropic images with an isovoxel size of 0.75 mm, using cubic and nearest neighbor interpolations, respectively [23]. The volumes of intereste (VOI) were cropped based on the center of volume of the circumscribed square of the maximum dimension annotations from the annotation data sets. The cropped VOI images were resized to 64 × 64 × 8–58 pixels to adjust the input image sizes to the NiftyNet platform [18]. A Laplacian of Gaussian (LoG) filter was applied to the interpolated CT images to reduce noise and enhance edges. The LoG was defined as

\(\text{L}\text{o}\text{G}(x, y,z)= -\frac{1}{\pi {\sigma }^{4}}\left[1-\frac{{x}^{2}+{y}^{2}+{z}^{2}}{2{\sigma }^{2}}\right]{e}^{-\frac{{x}^{2}+{y}^{2}+{z}^{2}}{2{\sigma }^{2}}} ,\)	(1)
where \(\sigma\) denotes the SD, which was set as \(\sigma =2.0\) in this study.

Two-step transfer learning based on Dense V-net

Figure 2 shows the architecture of the TSTL based on Dense V-net. In the first transfer learning, the weights of the lung cancer segmentation model were updated by inputting the poorly differentiated HCC data to construct the poorly differentiated HCC segmentation model. In the second transfer learning, the weights of the poorly differentiated HCC segmentation model were updated by feeding the well-differentiated HCC data to construct the well-differentiated HCC segmentation model.

The reason why we employed the Dense V-net was that Cui et al. [18] reported that Dense V-net was feasible for segmentation of lung cancer regions, which achieved a 3D-DSC of 0.832 ± 0.072 and HD of 4.57 ± 2.44 mm. Dense V-net is composed of two advantageous features, the dense connection, and V-network. The main advantage of dense connection is that it inputs all the features extracted in the previous layer, avoiding overfitting when using deeper networks and smaller data sets. The V-network is an extension of the U-network, where the input image is changed to a 3D image and the residual learning of the U-Net is employed to improve the performance.

Transfer learning is a technique for adapting a model learned in one domain to another domain. Deep convolutional activation features learned from ImageNet, a large natural image database, have been successfully transferred to differentiate patients into the degree of differentiation for HCCs from CT images, with little training data [24]. We hypothesized that a DL model for lung cancer would be applicable to segmentation of poorly differentiated HCC, which could then be retrained for well-differentiated HCC, because of the more similarity between lung cancer and poorly differentiated HCC, as well as that between poorly and well-differentiated HCCs, than natural images. We verified that lung tumors for 12 cases, which were randomly selected, were similar to HCC for different poorly differentiated 12 cases by making all combinations between two datasets, whose result was an average cross-correlation coefficient of 0.719 ± 0.12.

Three types of DCE CT images at arterial, equilibrium, and portal phases were fed into three channels of Dense V-net, respectively, and HCC contours (reference contours) were given as annotation data to the network. Optimal hyperparameters for building a model with weights for a poorly differentiated HCC were as follows: an activation function of tanh; a loss function of Dice loss [25]; a dropout rate of 60; an optimization algorithm of RMSProp [26]; a batch size of 8; a learning rate of 0.001; and the number of epochs of 50. Furthermore, the model with weights for a well-differentiated HCC was as follows: tanh; Dice loss; 65; RMSProp; 8; 0.001; and 60. It is advantageous to employ data augmentation techniques such as image rotation, flipping, and scaling to reduce the DL overfitting problem using smaller amounts of medical image data [27]. The images were randomly rotated from − 40° to + 40°, randomly flipped, and randomly scaled from − 20% to + 20%.

Evaluation of proposed models

The proposed models were evaluated with the nested cross-validation method, because Dora et al. [21] showed that nested cross-validation could avoid an overfitting problem and evaluate the generalization performance of models. In the nested cross-validation, a cross-validation was divided into two parts: inner and outer cross-validation loops. Figure 3 shows a procedure of the nested cross-validation method. The whole dataset is split into 4 folds, one outer fold is retained as a 1st test dataset, and the remaining 3 outer folds are merged and split into 4 folds for an inner cross-validation. Hyperparameters were optimized in the 1st inner training, and the constructed model was evaluated with the 1st outer test dataset in the outer cross-validation. This procedure is repeated for the rest of the outer training and tests. The model with the highest 3D-DSC in the four outer folds for poorly differentiated HCC is employed in the second transfer learning. To investigate the impact of TSTL on the performance of the proposed segmentation models, all evaluation indices were obtained with and without the TSTL.

3D-DSC and the 95% HD were used to evaluate the accuracy of the proposed segmentation models [19, 20]. The 3D-DSCs denote the degree of region similarity between reference HCC regions and the regions predicted using the proposed model [19]. The HD is a mathematical construct used to measure the 3D closeness of two sets of points that are subsets of a metric space. If two images are identical, the HD is zero. The larger HD becomes, the more the shapes of two images differ. In this study, we used 95% HD, that is, the 95th-percentile of the HD between two sets, because the maximum HD is highly sensitive to outliers [20].

Since the longest HCC diameters are assessed in clinical practice, Spearman's rank correlation coefficients (SCs) and the errors in the longest HCC diameters were calculated between the annotated and predicted regions in poorly and well-differentiated HCC. The longest HCC diameters of the regions were measured using an in-house program based on principal component analysis (PCA) (MATLAB version 2019b, MathWorks Inc., Natick, MA) [28]. For the measurement of HCC diameters, the coordinates of voxels within an estimated tumor region were extracted to calculate a covariance matrix. The eigenvalues were calculated by applying a singular value decomposition to the covariance matrix. The longest HCC diameters were estimated using the square roots of the first eigenvalues. The average, minimum, maximum, and percentage errors for the longest HCC diameters were evaluated. Furthermore, to evaluate the impact of the HCC diameter on the accuracy of region segmentation by 3D-DSC, the dataset was divided into two groups according to the median of the longest HCC diameter of the annotations.

Two Dense V-nets were built on the NiftyNet platform [29]. NiftyNet is a convolutional neural network platform designed for medical image analysis research and is an open-source system that uses the TensorFlow framework (version: 1.14.0). NiftyNet was implemented in Python for TensorFlow to improve the readability of the code. All calculations were performed using a GPU of NVIDIA RTX2080 (NVIDIA Corporation, Los Alamitos, CA).

Statistical analysis

R (version 4.0.3) was used for all statistical analyses [30]. In the model for poorly and well-differentiated HCC, the Mann‒Whitney U-test was applied to assess statistically significant differences in the average 3D-DSC and 95% HD between the proposed model and the model without TSTL.

Figure 4 shows the average 3D-DSCs obtained by proposed models with TSTL and models without TSTL for lung cancer, respectively. The models without TSTL were independently constructed for poorly and well-differentiated HCCs from scratch. For poorly differentiated HCC, the average 3D-DSCs were 0.849 ± 0.078 by the proposed model and 0.684 ± 0.154 by the model without the first TL (p < 0.05). Regarding well-differentiated HCC, the average 3D-DSCs were 0.811 ± 0.089 by the proposed model and 0.700 ± 0.162 (p < 0.05) by the model without the second TL. Figure 5 illustrates regions (green line) that were well-predicted by the proposed model for two patients with poorly differentiated HCC (Patients 1 and 2), and two patients with well-differentiated HCC (Patients 3 and 4). The contrast-injection phase of the HCC image data was the arterial phase. Images in this figure are arranged at ± 2 slices from the center slice of the tumor. Figure 6 depicts the predicted regions (green line) with the lowest 3D-DSC values (0.615 and 0.605, respectively) for Patient 5 with poorly and Patient 6 with well-differentiated HCC.

Figure 7 shows the 95% HDs obtained by the proposed models using TSTL and the models without TSTL. The average 95% HDs for poorly and well-differentiated HCC using the proposed model with TSTL were 1.98 ± 0.71 mm and 2.01 ± 0.83 mm, respectively, whereas those using the model without TSTL were 2.67 ± 1.04 mm and 2.78 ± 1.50 mm, respectively (p < 0.05).

Figures 8 and 9 show correlations in the longest HCC diameter between the predicted regions and annotations for poorly and well-differentiated HCC, respectively. Spearman’s rank correlation coefficients for the proposed model and the model without TSTL were 0.95 and 0.78, respectively, for poorly differentiated HCC, and 0.94 and 0.71, respectively, for well-differentiated HCC.

Figures 4, 7, 8, and 9 indicate the feasibility of using TSTL. Based on these data, the following two hypotheses were proven: (1) Separate segmentations of poorly and well-differentiated HCC regions could improve their accuracy, and (2) a DL model for lung cancer is applicable to segmentation of poorly differentiated HCC, and could be retrained for segmenting well-differentiated HCC, because of their similarity.

Figure 10 shows the average 3D-DSCs for patient groups with short and long HCC diameters. The patients were divided into the two groups (short and long diameter groups), according to the median of the longest annotated diameter for poorly and well-differentiated HCC. The median diameters for poorly and well-differentiated HCC were 18.4 mm and 17.3 mm, respectively. For poorly differentiated HCC, the average 3D-DSCs were 0.822 ± 0.061 in the short group and 0.875 ± 0.050 in the long group (p < 0.05). For well-differentiated HCC, the average 3D-DSCs were 0.800 ± 0.076 in the short group and 0.822 ± 0.081 in the long group (p > 0.05).

Table 2 shows the errors in the longest HCC diameters of the annotated image and the proposed segmentation models with and without TSTL. The average errors in the longest HCC diameter for poorly differentiated HCC using the proposed model with TSTL and the model without TSTL were 2.42 mm and 4.05 mm (p < 0.05), respectively. Those for well-differentiated HCC were 2.21 mm and 3.94 mm (p < 0.05), respectively.

Table 2

Errors in longest HCC diameters of annotations and the proposed models.
	Poorly differentiated HCC			Well-differentiated HCC
	Annotations	Proposed model	Model w/o 1st-TL	Annotations	Proposed model	Model w/o 2nd-TL
Average longest diameter (mm)	20.9	19.8	22.4	19.2	20.3	21.9
Average error (mm)	-	2.42	4.05	-	2.21	3.94
Minimum error (mm)	-	0.10	0.16	-	0.38	0.24
Maximum error (mm)	-	9.94	29.82	-	9.29	15.84
Percentage error (%)	-	12.2	18.1	-	10.9	18.0

Table 3 shows a comparison of the proposed model with three automated models developed in previous studies using the LiTS Challenge dataset [16]. This comparison is discussed in the next section.

Table 3

Comparison of Dice’s similarity coefficient (DSC) and Hausdorff distance (HD), 95% HD indices obtained from the present study and past studies. pHCC: poorly differentiated HCC, wHCC: well-differentiated HCC
	Number of cases	Data type	Evaluation method	Dataset	DSC	95% HD (mm)	HD (mm)
Present study	128 (pHCC: 80) (wHCC: 48)	Clinical data	Nested 4-fold cross validation	Training:96, Testing:32	All HCC: 0.834 (pHCC: 0.849) (wHCC: 0811)	All HCC: 1.99 (pHCC: 1.98) (wHCC: 2.01)	All HCC: 2.46 (pHCC: 2.57) (wHCC: 2.39)
Yading Yuan ¹² (2017)	200	Open data (LiTS)	Training & test	Training:130, Testing:70	0.820	-	6.26
Xiaomeng Li ¹³ (2018)	200	Open data (LiTS)	Training & test	Training:130, Testing:70	0.824	-	6.05
Abdul Qayyum ¹⁴ (2020)	200	Open data (LiTS)	Training & test	Training:130, Testing:70	0.825	-	6.10
Omar Ibrahim Alirr ¹⁵ (2020)	200	Open data (LiTS)	Training & test	Training:130, Testing:70	0.761	-	-

In this study, we have developed dual segmentation models for poorly and well-differentiated HCC, using TSTL, based on DCE CT images of lung cancer patients, as a poorly differentiated HCC model in the first step, followed by construction of a well-differentiated HCC model from the poorly differentiated HCC model in the second step.

The proposed models achieved a higher average 3D-DSC of 0.834 (0.849 and 0.811 for poorly and well-differentiated HCCs, respectively) than the average 3D-DSC of 0.820 to 0.825 obtained for unclassified HCC in previous studies (Table 3) [12–15]. The average HD of the proposed model was 2.46 mm (2.57 and 2.39 mm for poorly and well-differentiated HCCs, respectively), which were lower than the values of 6.10 to 6.26 mm obtained in past studies.

The LiTS challenge dataset used in previous studies is not categorized into poorly differentiated and well-differentiated HCCs [12–15]. On the other hand, we employed a categorized dataset, including poorly and well-differentiated HCC cases, and separately and consecutively constructed the proposed models for poorly and well-differentiated HCC using TSTL. There were two reasons for this approach. First, poorly and well-differentiated HCCs have different imaging traits on DCE CT images in the arterial, portal, and equilibrium phases, from a clinical point of view. Second, we hypothesized that the separate and consecutive constructions (i.e., TSTL) for poorly and well-differentiated HCCs would facilitate improved segmentation accuracy. Consequently, the segmentation accuracy (3D-DSC and 95% HD) and measurement accuracy of the longest diameter (Spearman’s rank correlation coefficient and error) achieved by our models were superior to those obtained in previous HCC segmentation studies [12–15].

As shown in Fig. 4 and Table 2, there are cases with low 3D-DSCs and large errors, such as the two cases (Fig. 6) with the lowest 3D-DSC for poorly and well-differentiated HCCs. One of the causes for this could be smaller regions, which caused lower segmentation accuracies, and particularly overestimation of poorly differentiated HCC regions (Figs. 8(a) and 10(a)). The other causes may be low-contrast and blurred regions. Therefore, the HCC regions may have been segmented jointly with angiogenic vessels. The issues of small size and low contrast or blurring should be addressed in future studies.

This study had two limitations. First, we did not address moderately differentiated or undifferentiated HCCs, but rather focused on poorly and well-differentiated HCCs. Therefore, the proposed models should be applied to moderately differentiated or undifferentiated HCCs in future studies. The second limitation is that the proposed models have not been applied to large databases. This should be done by collecting clinical data from our hospital as well as other institutions in future, to evaluate the generalizability of these models.

The proposed models constructed by TSTL from a lung cancer dataset showed the potential for automatic segmentation of poorly and well-differentiated HCC regions on the DCE CT images. The average 3D-DSC and 95% HD values obtained by the proposed models are 0.834 and 1.99 mm, respectively. Therefore, the proposed model has the potential to be useful in delineating HCCs as a treatment option and in assisting hepatologists and radiologists.

Acknowledgments

This study was partially supported by a grant from Center for Clinical and Translational Research of Kyushu University Hospital and JSPS KAKENHI Grant Number JP20K08084. The authors are grateful to all members of the Arimura Laboratory (http://web.shs.kyushu-u.a c.jp/~arimura/), Department of radiology, Saga University Hospital for their valuable comments and helpful discussion and to Hirokazu Takahashi, Satoshi Oeda and Hiroshi Isoda, Liver center, Saga University Hospital for their comments for clinical practice of HCC. We would like to thank Editage (www.editage.com) for English language editing.

Author Contributions

Conceptualization: Hidetaka Arimura, Noriyuki Nagami

Data curation: Hidetaka Arimura, Noriyuki Nagami, Manabu Ogata, Junichi Nojiri

Formal analysis: Noriyuki Nagami

Funding acquisition: Hidetaka Arimura

Investigation: Noriyuki Nagami, Hidetaka Arimura

Methodology: Noriyuki Nagami, Hidetaka Arimura

Project administration: Hidetaka Arimura,

Resources: Noriyuki Nagami, Hidetaka Arimura, Junichi Nojiri, Cui Yunhao, Kenta Ninomiya, Manabu Ogata, Shigetoshi Kitamura, Hiroyuki Irie

Software: Noriyuki Nagami, Kenta Ninomiya, Cui Yunhao

Supervision: Hidetaka Arimura, Shigetoshi Kitamura, Hiroyuki Irie

Validation: Noriyuki Nagami, Hidetaka Arimura

Visualization: Noriyuki Nagami, Hidetaka Arimura

Writing – original draft: Noriyuki Nagami, Hidetaka Arimura

Sung H, Ferlay J, Siegel R, Laversanne M, Soerjomataram I, Jemal A, Bray F (2021) Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J. Clin. 71(3): 209-249. https://doi.org/10.3322/caac.21660
Villanueva A (2019) Hepatocellular carcinoma. N Engl J Med 380: 1450–1462. https://doi.org/10.1056/nejmra1713263
World Health Organization Liver Factsheet. Globocan Available at: https://gco.iarc.fr/today/data/factsheets/cancers/11-Liver-fact-sheet.pdf. Accessed Jun 11, 2022.
Kim E, Viatour P (2020) Hepatocellular carcinoma: old friends and new tricks. Exp Mol Med 52: 1898-1907. https://doi.org/10.1038/s12276-020-00527-1
The Japan society of hepatology (2017) Clinical Practice Guidelines for Hepatocellular Carcinoma 2017. Kanehara & Co., LTD, Tokyo.
Tarhan NC, Hatipoğlu T, Ercan E, Bener M, Keleş G, Başaran C et al (2011) Correlation of dynamic multidetector CT findings with pathological grades of hepatocellular carcinoma. Diagn Interv Radiol 17(4):328–333. https://doi.org/10.4261/1305-3825.dir.2682-09.3
Chen W, Zhang T, Xu L, Zhao L, Liu H, Gu LR et al (2021) Radiomics analysis of contrast-enhanced CT for hepatocellular carcinoma grading. Front Oncol 11:660509. https://doi.org/10.3389/fonc.2021.660509
Pérez-Saborido B, de los Galanes SJ, Menéu-Dı́az JC, Romero CJ, Elola-Olaso AM, Suárez YF et al (2007) Tumor recurrence after liver transplantation for hepatocellular carcinoma: recurrence pathway and prognostic factors. Transplant Proc 39(7):2304–2307. https://doi.org/10.1016/j.transproceed.2007.06.059
Shinkawa H, Tanaka S, Kabata D, Takemura S, Amano R, Kimura K et al (2021) The prognostic impact of tumor differentiation on recurrence and survival after resection of hepatocellular carcinoma is dependent on tumor size. Liver cancer 10(5): 461-472. https://doi.org/10.1159/000517992
Kim Y S, Kim J W, Yoon W S, Kang M K, Lee I K, Kim T H et al (2016) Interobserver variability in gross tumor volume delineation for hepatocellular carcinoma. Strahlenther Onkol 192: 714-721. https://doi.org/10.1007/s00066-016-1028-2
Yasaka K, Akai H, Abe O, Kiryu S (2018) Deep learning with CNN showed high diagnostic performance in differentiation of liver masses at dynamic CT. Radiology 286(3-March): 887-896. https://doi.org/10.1148/radiol.2017170706
Yuan Y (2017) Hierarchical convolutional-deconvolutional neural networks for automatic liver and tumor segmentation. arXiv preprint arXiv:1710.04540.
Li X, Chen H, Qi X, Dou Q, Fu C W, Heng P A (2018) H-Dense UNet: Hybrid densely connected UNet for liver and tumor segmentation from CT Volumes. IEEE Trans Med Imaging 37(12): 2663-2674.
Qayyum A, Lalande A, Meriaudeau F (2020) Automatic segmentation of tumors and affected organs in the abdomen using a 3D hybrid model for computed tomography imaging. Comput Biol Med 2020; 127: 104097. https://doi.org/10.1016/j.compbiomed.2020.104097
Alirr OI (2020) Deep learning and level set approach for liver and tumor segmentation from CT scans. J Appl Clin Med Phys 21(10): 200-209. https://doi.org/10.1002/acm2.13003
Bilic P, Christ P F, Vorontsov E, Chlebus G, Chen H, Dou Q et al (2019) The Liver Tumor Segmentation Benchmark (LiTS). arXiv preprint arXiv:1901.04056.
Zhang W, Yang R, Liang F, Liu G, Chen A, Wu H et al (2021) Prediction of microvascular invasion in hepatocellular carcinoma with a multi-disciplinary team-like radiomics fusion model on dynamic contrast-enhanced computed tomography. Front Oncol Mar; 11:1-11. https://doi.org/10.3389/fonc.2021.660629
Cui Y, Arimura H, Nakano R, Yoshitake T, Shioyama Y, Yabuuchi H (2021) Automated approach for segmenting gross tumor volumes for lung cancer stereotactic body radiation therapy using CT-based dense V-networks. J Radiat Res 62(2): 346-355. https://doi.org/10.1093/jrr/rraa132
Dice LR (1945) Measures of the amount of ecologic association between species. Ecology 26(3): 297-302.
Chen X, Sun S, Bai N, Han K, Liu Q, Yao S et al (2021) A deep learning-based auto-segmentation system for organs-at-risk on whole-body computed tomography images for radiation therapy. Radiother Oncol 160: 175-184. https://doi.org/10.1016/j.radonc.2021.04.019
Dora L, Agrawal S, Panda R, Abraham A (2018) Nested cross-validation based adaptive sparse representation algorithm and its application to pathological brain classification. Expert Syst Appl 114: 313-321. https://doi.org/10.1016/j.eswa.2018.07.039
D Slicer. https://www.slicer.org/
Smith D J (2018) A nearest-neighbour discretisation of the regularized stokeslet boundary integral equation. J Comput Phys 358: 88–102. https://doi.org/10.1016/j.jcp.2017.12.008
Yamada A, Oyama K, Fujita S, Yoshizawa E, Ichinohe F, Komatsu D et al (2019) Dynamic contrast-enhanced computed tomography diagnosis of primary liver cancers using transfer learning of pretrained convolutional neural networks: Is registration of multiphasic images necessary? Int J Comput Assist Radiol Surg 14(8): 1295-1301. https://doi.org/10.1007/s11548-019-01987-1
Milletari F, Navab N, Ahmadi S A (2016) V-Net: Fully convolutional neural networks for volumetric medical image segmentation. International Conference on 3D Vision IEEE Stanford, CA, 565-571.
Tieleman T, Hinton G (2012) Lecture 6.5-RMSProp: Divide the gradient by a running average of its recent magnitude. COURSERA: Neural networks for machine learning 4(2): 26–31.
Chlap P, Min H, Vandenberg N, Dowling J, Holloway L, Haworth A (2021) A review of medical image data augmentation techniques for deep learning applications. J Med Imaging Radiat Oncol 65: 545-563. https://doi.org/10.1111/1754-9485.13261
Rodarmel C, Shan J (2002) Principal component analysis for hyperspectral image classification. Surv L Inf Sci 62(2): 115-122.
Gibson E, Li W, Sudre C, Fidon L, Shakir D I, Wang G et al (2018) NiffyNet: A deep-learning platform for medical imaging. Comput Methods Programs Biomed 158: 113–122. https://dx.doi.org/10.1016%2Fj.cmpb.2018.01.025
R. https://www.r-project.org/

Download PDF

Reviewers agreed at journal
22 Jun, 2022
Reviewers invited by journal
21 Jun, 2022
Editor invited by journal
21 Jun, 2022
Editor assigned by journal
20 Jun, 2022
First submitted to journal
18 Jun, 2022

You are reading this latest preprint version

Dual segmentation models for poorly and well-differentiated hepatocellular carcinoma using two-step transfer deep learning on dynamic contrast-enhanced CT images

Status:

Version 1

Abstract

Figures

Introduction

Materials And Methods

Clinical cases

Preprocessing of HCC region datasets for CT images

Two-step transfer learning based on Dense V-net

Evaluation of proposed models

Statistical analysis

Results

Discussion

Conclusion

Declarations

References

Status:

Version 1