1
|
Wang D, Liu C, Wang X, Liu X, Lan C, Zhao P, Cho WC, Graeber MB, Liu Y. Automated Machine-Learning Framework Integrating Histopathological and Radiological Information for Predicting IDH1 Mutation Status in Glioma. FRONTIERS IN BIOINFORMATICS 2021; 1:718697. [PMID: 36303770 PMCID: PMC9581043 DOI: 10.3389/fbinf.2021.718697] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [Grants] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 06/01/2021] [Accepted: 09/28/2021] [Indexed: 09/01/2023] Open
Abstract
Diffuse gliomas are the most common malignant primary brain tumors. Identification of isocitrate dehydrogenase 1 (IDH1) mutations aids the diagnostic classification of these tumors and the prediction of their clinical outcomes. While histology continues to play a key role in frozen section diagnosis, as a diagnostic reference and as a method for monitoring disease progression, recent research has demonstrated the ability of multi-parametric magnetic resonance imaging (MRI) sequences for predicting IDH genotypes. In this paper, we aim to improve the prediction accuracy of IDH1 genotypes by integrating multi-modal imaging information from digitized histopathological data derived from routine histological slide scans and the MRI sequences including T1-contrast (T1) and Fluid-attenuated inversion recovery imaging (T2-FLAIR). In this research, we have established an automated framework to process, analyze and integrate the histopathological and radiological information from high-resolution pathology slides and multi-sequence MRI scans. Our machine-learning framework comprehensively computed multi-level information including molecular level, cellular level, and texture level information to reflect predictive IDH genotypes. Firstly, an automated pre-processing was developed to select the regions of interest (ROIs) from pathology slides. Secondly, to interactively fuse the multimodal complementary information, comprehensive feature information was extracted from the pathology ROIs and segmented tumor regions (enhanced tumor, edema and non-enhanced tumor) from MRI sequences. Thirdly, a Random Forest (RF)-based algorithm was employed to identify and quantitatively characterize histopathological and radiological imaging origins, respectively. Finally, we integrated multi-modal imaging features with a machine-learning algorithm and tested the performance of the framework for IDH1 genotyping, we also provided visual and statistical explanation to support the understanding on prediction outcomes. The training and testing experiments on 217 pathologically verified IDH1 genotyped glioma cases from multi-resource validated that our fully automated machine-learning model predicted IDH1 genotypes with greater accuracy and reliability than models that were based on radiological imaging data only. The accuracy of IDH1 genotype prediction was 0.90 compared to 0.82 for radiomic result. Thus, the integration of multi-parametric imaging features for automated analysis of cross-modal biomedical data improved the prediction accuracy of glioma IDH1 genotypes.
Collapse
Affiliation(s)
- Dingqian Wang
- School of Computer Science, The University of Sydney, Sydney, NSW, Australia
| | - Cuicui Liu
- Department of Neurology, Shandong Provincial Hospital Affiliated to Shandong First Medical University, Jinan, China
| | - Xiuying Wang
- School of Computer Science, The University of Sydney, Sydney, NSW, Australia
| | - Xuejun Liu
- Department of Radiology, Hospital Affiliated to Qingdao University, Qingdao, China
| | - Chuanjin Lan
- Department of Neurosurgery, Shandong Provincial Hospital Affiliated to Shandong First Medical University, Jinan, China
- Department of Neurosurgery, Shandong Provincial Hospital Affiliated to Shandong University, Jinan, China
| | - Peng Zhao
- Department of Neurosurgery, Shandong Provincial Hospital Affiliated to Shandong First Medical University, Jinan, China
- Department of Neurosurgery, Shandong Provincial Hospital Affiliated to Shandong University, Jinan, China
| | - William C. Cho
- Department of Clinical Oncology, Queen Elizabeth Hospital, Kowloon, Hong Kong, SAR China
| | - Manuel B. Graeber
- Ken Parker Brain Tumor Research Laboratories, Brain and Mind Centre, Faculty of Medicine and Health, The University of Sydney, Sydney, NSW, Australia
| | - Yingchao Liu
- Department of Neurosurgery, Shandong Provincial Hospital Affiliated to Shandong First Medical University, Jinan, China
- Department of Neurosurgery, Shandong Provincial Hospital Affiliated to Shandong University, Jinan, China
| |
Collapse
|
2
|
Wang X, Wang D, Yao Z, Xin B, Wang B, Lan C, Qin Y, Xu S, He D, Liu Y. Machine Learning Models for Multiparametric Glioma Grading With Quantitative Result Interpretations. Front Neurosci 2019; 12:1046. [PMID: 30686996 PMCID: PMC6337068 DOI: 10.3389/fnins.2018.01046] [Citation(s) in RCA: 32] [Impact Index Per Article: 6.4] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 10/24/2018] [Accepted: 12/24/2018] [Indexed: 12/11/2022] Open
Abstract
Gliomas are the most common primary malignant brain tumors in adults. Accurate grading is crucial as therapeutic strategies are often disparate for different grades and may influence patient prognosis. This study aims to provide an automated glioma grading platform on the basis of machine learning models. In this paper, we investigate contributions of multi-parameters from multimodal data including imaging parameters or features from the Whole Slide images (WSI) and the proliferation marker Ki-67 for automated brain tumor grading. For each WSI, we extract both visual parameters such as morphology parameters and sub-visual parameters including first-order and second-order features. On the basis of machine learning models, our platform classifies gliomas into grades II, III, and IV. Furthermore, we quantitatively interpret and reveal the important parameters contributing to grading with the Local Interpretable Model-Agnostic Explanations (LIME) algorithm. The quantitative analysis and explanation may assist clinicians to better understand the disease and accordingly to choose optimal treatments for improving clinical outcomes. The performance of our grading model was evaluated with cross-validation, which randomly divided the patients into non-overlapping training and testing sets and repeatedly validated the model on the different testing sets. The primary results indicated that this modular platform approach achieved the highest grading accuracy of 0.90 ± 0.04 with support vector machine (SVM) algorithm, with grading accuracies of 0.91 ± 0.08, 0.90 ± 0.08, and 0.90 ± 0.07 for grade II, III, and IV gliomas, respectively.
Collapse
Affiliation(s)
- Xiuying Wang
- School of Information Technologies, The University of Sydney, Sydney, NSW, Australia
| | - Dingqian Wang
- School of Information Technologies, The University of Sydney, Sydney, NSW, Australia
| | - Zhigang Yao
- Department of Pathology, Provincial Hospital Affiliated to Shandong University, Jinan, China
| | - Bowen Xin
- School of Information Technologies, The University of Sydney, Sydney, NSW, Australia
| | - Bao Wang
- School of Medicine, Shandong University, Jinan, China
| | - Chuanjin Lan
- School of Medicine, Shandong University, Jinan, China
| | - Yejun Qin
- Department of Pathology, Provincial Hospital Affiliated to Shandong University, Jinan, China
| | - Shangchen Xu
- Department of Neurosurgery, Provincial Hospital Affiliated to Shandong University, Jinan, China
| | - Dazhong He
- School of Information and Communication Engineering, Beijing University of Posts and Telecommunications, Beijing, China
| | - Yingchao Liu
- Department of Neurosurgery, Provincial Hospital Affiliated to Shandong University, Jinan, China
| |
Collapse
|
3
|
Delgado AF, Fahlström M, Nilsson M, Berntsson SG, Zetterling M, Libard S, Alafuzoff I, van Westen D, Lätt J, Smits A, Larsson EM. Diffusion Kurtosis Imaging of Gliomas Grades II and III - A Study of Perilesional Tumor Infiltration, Tumor Grades and Subtypes at Clinical Presentation. Radiol Oncol 2017; 51:121-129. [PMID: 28740446 PMCID: PMC5514651 DOI: 10.1515/raon-2017-0010] [Citation(s) in RCA: 28] [Impact Index Per Article: 4.0] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 08/15/2016] [Accepted: 01/08/2017] [Indexed: 12/12/2022] Open
Abstract
BACKGROUND Diffusion kurtosis imaging (DKI) allows for assessment of diffusion influenced by microcellular structures. We analyzed DKI in suspected low-grade gliomas prior to histopathological diagnosis. The aim was to investigate if diffusion parameters in the perilesional normal-appearing white matter (NAWM) differed from contralesional white matter, and to investigate differences between glioma malignancy grades II and III and glioma subtypes (astrocytomas and oligodendrogliomas). PATIENTS AND METHODS Forty-eight patients with suspected low-grade glioma were prospectively recruited to this institutional review board-approved study and investigated with preoperative DKI at 3T after written informed consent. Patients with histologically proven glioma grades II or III were further analyzed (n=35). Regions of interest (ROIs) were delineated on T2FLAIR images and co-registered to diffusion MRI parameter maps. Mean DKI data were compared between perilesional and contralesional NAWM (student's t-test for dependent samples, Wilcoxon matched pairs test). Histogram DKI data were compared between glioma types and glioma grades (multiple comparisons of mean ranks for all groups). The discriminating potential for DKI in assessing glioma type and grade was assessed with receiver operating characteristics (ROC) curves. RESULTS There were significant differences in all mean DKI variables between perilesional and contralesional NAWM (p=<0.000), except for axial kurtosis (p=0.099). Forty-four histogram variables differed significantly between glioma grades II (n=23) and III (n=12) (p=0.003-0.048) and 10 variables differed significantly between ACs (n=18) and ODs (n=17) (p=0.011-0.050). ROC curves of the best discriminating variables had an area under the curve (AUC) of 0.657-0.815. CONCLUSIONS Mean DKI variables in perilesional NAWM differ significantly from contralesional NAWM, suggesting altered microstructure by tumor infiltration not depicted on morphological MRI. Histogram analysis of DKI data identifies differences between glioma grades and subtypes.
Collapse
Affiliation(s)
- Anna F Delgado
- Department of Surgical Sciences, Radiology, Uppsala University, Uppsala, Sweden.,Department of Clinical Neuroscience, Karolinska Institute, Stockholm, Sweden
| | - Markus Fahlström
- Department of Surgical Sciences, Radiology, Uppsala University, Uppsala, Sweden
| | | | - Shala G Berntsson
- Department of Neuroscience, Neurology, Uppsala University, Uppsala, Sweden
| | - Maria Zetterling
- Department of Neuroscience, Neurosurgery, Uppsala University, Uppsala, Sweden
| | - Sylwia Libard
- Department of Immunology, Genetics and Pathology, Section of pathology, Uppsala University Hospital and Uppsala University, Uppsala, Sweden
| | - Irina Alafuzoff
- Department of Immunology, Genetics and Pathology, Section of pathology, Uppsala University Hospital and Uppsala University, Uppsala, Sweden
| | | | - Jimmy Lätt
- Department of Imaging and Function, Skåne University Healthcare, Lund, Sweden
| | - Anja Smits
- Department of Neuroscience, Neurology, Uppsala University, Uppsala, Sweden
| | - Elna-Marie Larsson
- Department of Surgical Sciences, Radiology, Uppsala University, Uppsala, Sweden
| |
Collapse
|
4
|
Oncogenic signaling is dominant to cell of origin and dictates astrocytic or oligodendroglial tumor development from oligodendrocyte precursor cells. J Neurosci 2015; 34:14644-51. [PMID: 25355217 DOI: 10.1523/jneurosci.2977-14.2014] [Citation(s) in RCA: 37] [Impact Index Per Article: 4.1] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 11/21/2022] Open
Abstract
Stem cells, believed to be the cellular origin of glioma, are able to generate gliomas, according to experimental studies. Here we investigated the potential and circumstances of more differentiated cells to generate glioma development. We and others have shown that oligodendrocyte precursor cells (OPCs) can also be the cell of origin for experimental oligodendroglial tumors. However, the question of whether OPCs have the capacity to initiate astrocytic gliomas remains unanswered. Astrocytic and oligodendroglial tumors represent the two most common groups of glioma and have been considered as distinct disease groups with putatively different origins. Here we show that mouse OPCs can give rise to both types of glioma given the right circumstances. We analyzed tumors induced by K-RAS and AKT and compared them to oligodendroglial platelet-derived growth factor B-induced tumors in Ctv-a mice with targeted deletions of Cdkn2a (p16(Ink4a-/-), p19(Arf-/-), Cdkn2a(-/-)). Our results showed that glioma can originate from OPCs through overexpression of K-RAS and AKT when combined with p19(Arf) loss, and these tumors displayed an astrocytic histology and high expression of astrocytic markers. We argue that OPCs have the potential to develop both astrocytic and oligodendroglial tumors given loss of p19(Arf), and that oncogenic signaling is dominant to cell of origin in determining glioma phenotype. Our mouse data are supported by the fact that human astrocytoma and oligodendroglioma display a high degree of overlap in global gene expression with no clear distinctions between the two diagnoses.
Collapse
|
5
|
Spindle cell oncocytomas and granular cell tumors of the pituitary are variants of pituicytoma. Am J Surg Pathol 2013; 37:1694-9. [PMID: 23887161 DOI: 10.1097/pas.0b013e31829723e7] [Citation(s) in RCA: 130] [Impact Index Per Article: 11.8] [Reference Citation Analysis] [Abstract] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 01/13/2023]
Abstract
Pituicytomas are neoplasms that arise from pituicytes, which are specialized glia of the posterior pituitary. Pituicytes have 5 ultrastructural variants: light, dark, granular, ependymal, and oncocytic. Granular cell tumors of the pituitary gland are thought to arise from granular pituicytes. Spindle cell oncocytomas are considered to arise from folliculostellate cells, which are sustentacular cells of the adenohypophysis. Recent data suggest that, whereas pituicytes and all 3 tumor types are positive for TTF-1, folliculostellate cells are negative for TTF-1. We investigated 7 spindle cell oncocytomas, 4 pituicytomas, and 3 granular cell tumors for their genetic (BRAF(V600E) mutation and BRAF-KIAA fusion), immunohistochemical (GFAP, vimentin, S100 protein, olig2, IDH1-R132H, NF, galectin-3, chromogranin-A, CD56, EMA, CAM5.2, CD68, TTF-1, and bcl-2), and ultrastructural features to refine their classification. All tumors had nuclear positivity for TTF-1 and were negative for CAM5.2, chromogranin-A, and NF. GFAP, vimentin, S100, galectin-3, EMA, and CD68 were variably positive in the majority of the 3 tumor groups. Olig2 was only positive in 1 pituicytoma. Whereas granular cell tumors were negative for bcl-2 and CD56, pituicytomas and spindle cell oncocytomas showed variable positivity. All tumors were negative with the IDH1-R132H mutation-specific antibody, and none had evidence of BRAF alterations (BRAF(V600E) mutation and BRAF-KIAA fusion). Diffuse TTF-1 expression in nontumorous pituicytes, pituicytomas, spindle cell oncocytomas, and granular cell tumors indicates a common pituicyte lineage. The ultrastructural variants of pituicytes are reflected in these 3 morphologic variants of tumors arising from these cells. We propose the terminology "oncocytic pituicytomas" and "granular cell pituicytomas" to refine the classification of these lesions.
Collapse
|
6
|
Hirose T, Ishizawa K, Shimada S. Utility of in situ demonstration of 1p loss and p53 overexpression in pathologic diagnosis of oligodendroglial tumors. Neuropathology 2011; 30:586-96. [PMID: 20408960 DOI: 10.1111/j.1440-1789.2010.01116.x] [Citation(s) in RCA: 7] [Impact Index Per Article: 0.5] [Reference Citation Analysis] [Abstract] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 11/29/2022]
Abstract
To improve the diagnostic accuracy of oligodendroglial tumors and to find more convenient parameters that could predict the cytogenetic status, oligodendroglial and astrocytic tumors were cytogenetically and immunohistochemically investigated. Materials included 22 oligodendroglial tumors (15 oligodendrogliomas and 7 oligoastrocytomas) and 20 astrocytic tumors. 1p loss was examined with the fluorescence in situ hybridization (FISH) method. Expression of GFAP, Olig2 and p53 was immunohistochemically investigated and co-localization of GFAP and Olig2 was evaluated on double-immunostained sections. Furthermore, TP53 mutation analyses were carried out on three oligodendroglial tumors showing p53 protein overexpression with a direct sequence analysis. Our FISH studies demonstrated 1p loss in 73% of oligodendroglial tumors (80% oligodendrogliomas and 57% oligoastrocytomas) and in only 10% of astrocytic tumors. There were no clear-cut morphologic differences between 1p-deleted and 1p-intact oligodendroglial tumors. GFAP and Olig2 were expressed in most oligodendroglial and astrocytic tumors, and their cellular localization was almost independent of each other. Overexpression of p53 was observed in five oligodendroglial tumors, all of which were 1p-intact. In comparison, 16 oligodendroglial tumors with 1p deletion showed no overexpression of p53. TP53 missense mutations were detected in three of the p53 overexpressed oligodendroglial tumors studied. Our results suggest that 1p loss is almost specific to oligodendroglial tumors. Although the prediction of 1p status based solely on the morphologic features seems to be difficult, the immunohistochemistry for p53 is a useful tool in that p53 overexpression is closely related to the 1p-intact status in oligodendroglial tumors.
Collapse
Affiliation(s)
- Takanori Hirose
- Department of Pathology, Saitama Medical School, Moroyama, Iruma-gun, Saitama, Japan.
| | | | | |
Collapse
|