A Comparison of Deep Learning and Pharmacokinetic Model Selection Methods in Segmentation of High-Grade Glioma

Abstract

Purpose: Glioma tumor segmentation is an essential step in clinical decision making. Recently, computer-aided methods have been widely used for rapid and accurate delineation of the tumor regions. Methods based on image feature extraction can be used as fast methods, while segmentation based on the physiology and pharmacokinetic of the tissues is more accurate. This study aims to compare the performance of tumor segmentation based on these two different methods.
 Materials and Methods: Nested Model Selection (NMS) based on Extended-Toft’s model was applied to 190 Dynamic Contrast-Enhanced MRI (DCE-MRI) slices acquired from 25 Glioblastoma Multiforme (GBM) patients in 70 time-points. A model with three pharmacokinetic parameters, Model 3, is usually assigned to tumor voxel based on the time-contrast concentration signal. We utilized Deep-Net as a CNN network, based on Deeplabv3+ and layers of pre-trained resnet18, which has been trained with 17288 T1-Contrast MRI slices with HGG brain tumor to predict the tumor region in our 190 DCE MRI T1 images. The NMS-based physiological tumor segmentation was considered as a reference to compare the results of tumor segmentation by Deep-Net. Dice, Jaccard, and overlay similarity coefficients were used to evaluate the tumor segmentation accuracy and reliability of the Deep tumor segmentation method.
 Results: The results showed a relatively high similarity coefficient (Dice coefficient: 0.73±0.15, Jaccard coefficient: 0.66±0.17, and overlay coefficient: 0.71±0.15) between deep learning tumor segmentation and the tumor region identified by the NMS method. The results indicate that the deep learning methods may be used as accurate and robust tumor segmentation.
 Conclusion: Deep learning-based segmentation can play a significant role to increase the segmentation accuracy in clinical application, if their training process is completely automatic and independent from human error.

Automatic segmentation of glioblastoma multiform brain tumor in MRI images: Using Deeplabv3+ with pre-trained Resnet18 weights

“…This property leads to a better localization and consequently affects the segmentation process. The acceptable results reported in previous studies and current work confirm the performance of the ASPP in segmentation tasks [8,18,22,37,39]. As aforementioned, Deep-Net is a deep neural network based on the Deeplabv3+ structure which employs Resnet18 as its main feature extractor (in the encoder part), with substantial modifications and optimization.…”

“…They reached a DSC of 72%, 79% and 87% for these tumor sub-regions. As can be understood from the Table 4 and the previously reported scores, the DSC of the Deep-Net in segmentation of ET, TC, and WT classes (78%, 85%, and 87%) show a promising result to ensure that the Deep-Net structure can be accurate and reliable enough to be used in glioblastoma segmentation studies [18]. Despite the acceptable performance of the Deep-Net, some defects are observed in the segmented maps in some slices of test data.…”

“…A combination of Deeplabv3 + as main structure and Resnet18 as its initial weights worked well in GBM tumor segmentation from T 1 ce and FLAIR MRI modalities images. Recently, our team showed [18] that the trained Deep-Net can perform well in GBM tumor segmentation from DCE-MR images too. Thus, it can be claimed that the high sensitivity (accuracy) and high precision (reliably reproducible) for all classes reveal that the trained Deep-Net in this work is accurate enough and maybe used as a reliable help for neuro-radiologists in clinical segmentation practice.…”

“…Deep neural networks are capable of automatically extracting the appropriate features from the image and then learning and recognizing the pattern of the tumors and normal tissues, so the deep-learning algorithms have been widely used for tumor segmentation in the last few years [16][17][18]. Convolutional Neural Network (CNN) structure is the most popular structure among the deep neural networks for segmentation tasks [2,4,11,18,19]. So far, the different dimensional CNN with 2D [4,20], 2.5D [21], or 3D structures, [22][23][24][25] were used in tumor segmentation.…”

Feasibility Study of Convolutional Long Short‑Term Memory Network for Pulmonary Movement Prediction in CT Images

“…The mean value and standard deviation of target registration error were 1.00±0.53 mm for ten 4DCT datasets and 1.59±1.58 mm for ten DIRLAB datasets [ 14 ]. It should be noted that the state of the art deep learning techniques was recently used in different studies, such as predicting future frames in stock market prediction [ 15 ], traffic accident prediction [ 16 ], text recognition [ 17 , 18 ], precipitation prediction [ 19 ], weather forecasting [ 20 ], ocean temperature [ 21 ], medical imaging [ 13 , 14 , 22 ], direction of slip detection [ 23 ], and travel demand prediction [ 24 ].…”