Classification of glioblastoma versus primary central nervous system lymphoma using convolutional neural networks

Abstract

A subset of primary central nervous system lymphomas (PCNSL) are difficult to distinguish from glioblastoma multiforme (GBM) on magnetic resonance imaging (MRI). We developed a convolutional neural network (CNN) to distinguish these tumors on contrast-enhanced T1-weighted images. Preoperative brain tumor MRIs were retrospectively collected among 320 patients with either GBM (n = 160) and PCNSL (n = 160) from two academic institutions. The individual images from these MRIs consisted of a training set (n = 1894 GBM and 1245 PCNSL), a validation set (n = 339 GBM; 202 PCNSL), and a testing set (99 GBM and 108 PCNSL). Three CNNs using the EfficientNetB4 architecture were evaluated. To increase the size of the training set and minimize overfitting, random flips and changes to color were performed on the training set. Our transfer learning approach (with image augmentation and 292 epochs) yielded an AUC of 0.94 (95% CI: 0.91–0.97) for GBM and an AUC of 0.95 (95% CI: 0.92–0.98) for PCNL. In the second case (not augmented and 137 epochs), the images were augmented prior to training. The area under the curve for GBM was 0.92 (95% CI: 0.88–0.96) for GBM and an AUC of 0.94 (95% CI: 0.91–0.97) for PCNSL. For the last case (augmented, Gaussian noise and 238 epochs) the AUC for GBM was 0.93 (95% CI: 0.89–0.96) and an AUC 0.93 (95% CI = 0.89–0.96) for PCNSL. Even with a relatively small dataset, our transfer learning approach demonstrated CNNs may provide accurate diagnostic information to assist radiologists in distinguishing PCNSL and GBM.

A Deep Learning Model for Preoperative Differentiation of Glioblastoma, Brain Metastasis and Primary Central Nervous System Lymphoma: A Pilot Study

“…As recent computer-based automated segmentation algorithms need to be clinically validated, manual segmentation is still the current gold- standard, showing overall satisfactory results at the cost of intensive work, task-induced fatigue, and extensive processing time. In a recent study, McAvoy et al ( 36 ) developed an EfficientNetB4 DNN with high classification performance for glioblastoma (accuracy: 0.94) vs PCNSL (accuracy: 0.95) on whole brain scan analysis with no prior image segmentation. The authors advocated the superiority of their model compared to previous machine learning studies, as the overall preprocessing effort was sharply reduced.…”

12 Plagues of AI in Healthcare: A Practical Guide to Current Issues With Using Machine Learning in a Medical Context

“…These models can often be seen with linear ( 58 ) and decision-tree based models ( 24 ), although a number of other applications are increasingly being developed ( 53 ). In fact, DL based networks make up the majority of the highly sought after radiological-AI applications for the medical field ( 1 ), such as the systems that can diagnose brain cancer during surgery. Such networks provide the enthusiasm for the recent large scale efforts in the field to improve the explainability of advanced ML techniques ( 59 ).…”

“…AI has been met with a surge of interest in the scientific and medical communities due to the increasing number of patients receiving healthcare services and the concomitant increases in complexity of data, which is now available, but often uninterpretable by humans alone. These technologies demonstrate the ability to identify malignant tumor cells on imaging during brain surgery ( 1 ), unravel novel diseases into explainable mechanisms of viral mutations for therapeutic design ( 2 ), predict the progression of neurodegenerative diseases to begin earlier treatments ( 3 ), and assist with the interpretation of vast amounts of genomic data to identify novel sequence patterns ( 4 ), among a number of many other medical applications. Ultimately, the applications of AI in medicine can generally be grouped into two bold promises for healthcare providers: (1) the ability to present larger amounts of interpretable information to augment clinical judgements while also (2) providing a more systematic view over data that decreases our biases.…”

Digital Pathology and Artificial Intelligence Applications in Pathology

“…In order to manage the patient at the appropriate time, it may be very difficult to carry out all the tests. Studies have reported that AI-based algorithms can predict the expression of biomarkers and genetic mutations of tumors by evaluating only the morphological characteristics of hematoxylin and eosin (H&E)-stained slides that are basically produced for pathological diagnosis [ 21 22 26 27 28 ]. Studies have also shown that novel biomarkers can be discovered by utilizing AI models in various types of cancers [ 29 30 31 ].…”

“…Recently, pathological AI models are being actively developed in the field of neurooncology, and they mainly target gliomas. In detail, pathological AI models being developed focus on the classification of gliomas, the mutation status of cancer-related genes, predicting patients’ prognosis, discovering new histological implicating features of tumors affecting cancer behaviors, and analyzing tumor microenvironment [ 27 28 32 33 34 ]. Most of the research uses H&E-stained slides, immunohistochemical stained slides, and, in some cases, genetic analysis information.…”