Deep Learning for Automatic Differential Diagnosis of Primary Central Nervous System Lymphoma and Glioblastoma: Multi‐Parametric Magnetic Resonance Imaging Based Convolutional Neural Network Model
Wei Xia,Bin Hu,Haiqing Li,Wei Shi,Ying Tang,Yang Yu,Chen Geng,Qiuwen Wu,Liqin Yang,Zekuan Yu,Daoying Geng,Yuxin Li
DOI: https://doi.org/10.1002/jmri.27592
IF: 4.4
2021-03-11
Journal of Magnetic Resonance Imaging
Abstract:<section class="article-section__content"><h3 class="article-section__sub-title section1"> Background</h3><p>Differential diagnosis of primary central nervous system lymphoma (PCNSL) and glioblastoma (GBM) is useful to guide treatment strategies.</p></section><section class="article-section__content"><h3 class="article-section__sub-title section1"> Purpose</h3><p>To investigate the use of a convolutional neural network (CNN) model for differentiation of PCNSL and GBM without tumor delineation.</p></section><section class="article-section__content"><h3 class="article-section__sub-title section1"> Study Type</h3><p>Retrospective.</p></section><section class="article-section__content"><h3 class="article-section__sub-title section1"> Population</h3><p>A total of 289 patients with PCNSL (136) or GBM (153) were included, the average age of the cohort was 54 years, and there were 173 men and 116 women.</p></section><section class="article-section__content"><h3 class="article-section__sub-title section1"> Field Strength/Sequence</h3><p>3.0 T Axial contrast‐enhanced T<sub>1</sub>‐weighted spin‐echo inversion recovery sequence (CE‐T<sub>1</sub>WI), T<sub>2</sub>‐weighted fluid‐attenuation inversion recovery sequence (FLAIR), and diffusion weighted imaging (DWI, <i>b</i> = 0 second/mm<sup>2</sup>, 1000 seconds/mm<sup>2</sup>). </p></section><section class="article-section__content"><h3 class="article-section__sub-title section1"> Assessment</h3><p>A single‐parametric CNN model was built using CE‐T<sub>1</sub>WI, FLAIR, and the apparent diffusion coefficient (ADC) map derived from DWI, respectively. A decision‐level fusion based multi‐parametric CNN model (DF‐CNN) was built by combining the predictions of single‐parametric CNN models through logistic regression. An image‐level fusion based multi‐parametric CNN model (IF‐CNN) was built using the integrated multi‐parametric MR images. The radiomics models were developed. The diagnoses by three radiologists with 6 years (junior radiologist Y.Y.), 11 years (intermediate‐level radiologist Y.T.), and 21 years (senior radiologist Y.L.) of experience were obtained. </p></section><section class="article-section__content"><h3 class="article-section__sub-title section1"> Statistical Analysis</h3><p>The 5‐fold cross validation was used for model evaluation. The Pearson's chi‐squared test was used to compare the accuracies. <i>U</i>‐test and Fisher's exact test were used to compare clinical characteristics. </p></section><section class="article-section__content"><h3 class="article-section__sub-title section1"> Results</h3><p>The CE‐T<sub>1</sub>WI, FLAIR, and ADC based single‐parametric CNN model had accuracy of 0.884, 0.782, and 0.700, respectively. The DF‐CNN model had an accuracy of 0.899 which was higher than the IF‐CNN model (0.830, <i>P</i> = 0.021), but had no significant difference in accuracy compared to the radiomics model (0.865, <i>P</i> = 0.255), and the senior radiologist (0.906, <i>P</i> = 0.886). </p></section><section class="article-section__content"><h3 class="article-section__sub-title section1"> Data Conclusion</h3><p>A CNN model can differentiate PCNSL from GBM without tumor delineation, and comparable to the radiomics models and radiologists.</p></section><section class="article-section__content"><h3 class="article-section__sub-title section1"> Level of Evidence</h3><p>4</p></section><section class="article-section__content"><h3 class="article-section__sub-title section1"> Technical Efficacy</h3><p>Stage 2</p></section>
radiology, nuclear medicine & medical imaging