Automated quantitative tumour response assessment of MRI in neuro-oncology with artificial neural networks: a multicentre, retrospective study.

BACKGROUND: The Response Assessment in Neuro-Oncology (RANO) criteria and requirements for a uniform protocol have been introduced to standardise assessment of MRI scans in both clinical trials and clinical practice. However, these criteria mainly rely on manual two-dimensional measurements of contrast-enhancing (CE) target lesions and thus restrict both reliability and accurate assessment of tumour burden and treatment response. We aimed to develop a framework relying on artificial neural networks (ANNs) for fully automated quantitative analysis of MRI in neuro-oncology to overcome the inherent limitations of manual assessment of tumour burden. METHODS: In this retrospective study, we compiled a single-institution dataset of MRI data from patients with brain tumours being treated at Heidelberg University Hospital (Heidelberg, Germany; Heidelberg training dataset) to develop and train an ANN for automated identification and volumetric segmentation of CE tumours and non-enhancing T2-signal abnormalities (NEs) on MRI. Independent testing and large-scale application of the ANN for tumour segmentation was done in a single-institution longitudinal testing dataset from the Heidelberg University Hospital and in a multi-institutional longitudinal testing dataset from the prospective randomised phase 2 and 3 European Organisation for Research and Treatment of Cancer (EORTC)-26101 trial (NCT01290939), acquired at 38 institutions across Europe. In both longitudinal datasets, spatial and temporal tumour volume dynamics were automatically quantified to calculate time to progression, which was compared with time to progression determined by RANO, both in terms of reliability and as a surrogate endpoint for predicting overall survival. We integrated this approach for fully automated quantitative analysis of MRI in neuro-oncology within an application-ready software infrastructure and applied it in a simulated clinical environment of patients with brain tumours from the Heidelberg University Hospital (Heidelberg simulation dataset). FINDINGS: For training of the ANN, MRI data were collected from 455 patients with brain tumours (one MRI per patient) being treated at Heidelberg hospital between July 29, 2009, and March 17, 2017 (Heidelberg training dataset). For independent testing of the ANN, an independent longitudinal dataset of 40 patients, with data from 239 MRI scans, was collected at Heidelberg University Hospital in parallel with the training dataset (Heidelberg test dataset), and 2034 MRI scans from 532 patients at 34 institutions collected between Oct 26, 2011, and Dec 3, 2015, in the EORTC-26101 study were of sufficient quality to be included in the EORTC-26101 test dataset. The ANN yielded excellent performance for accurate detection and segmentation of CE tumours and NE volumes in both longitudinal test datasets (median DICE coefficient for CE tumours 0·89 [95% CI 0·86-0·90], and for NEs 0·93 [0·92-0·94] in the Heidelberg test dataset; CE tumours 0·91 [0·90-0·92], NEs 0·93 [0·93-0·94] in the EORTC-26101 test dataset). Time to progression from quantitative ANN-based assessment of tumour response was a significantly better surrogate endpoint than central RANO assessment for predicting overall survival in the EORTC-26101 test dataset (hazard ratios ANN 2·59 [95% CI 1·86-3·60] vs central RANO 2·07 [1·46-2·92]; p<0·0001) and also yielded a 36% margin over RANO (p<0·0001) when comparing reliability values (ie, agreement in the quantitative volumetrically defined time to progression [based on radiologist ground truth vs automated assessment with ANN] of 87% [266 of 306 with sufficient data] compared with 51% [155 of 306] with local vs independent central RANO assessment). In the Heidelberg simulation dataset, which comprised 466 patients with brain tumours, with 595 MRI scans obtained between April 27, and Sept 17, 2018, automated on-demand processing of MRI scans and quantitative tumour response assessment within the simulated clinical environment required 10 min of computation time (average per scan). INTERPRETATION: Overall, we found that ANN enabled objective and automated assessment of tumour response in neuro-oncology at high throughput and could ultimately serve as a blueprint for the application of ANN in radiology to improve clinical decision making. Future research should focus on prospective validation within clinical trials and application for automated high-throughput imaging biomarker discovery and extension to other diseases. FUNDING: Medical Faculty Heidelberg Postdoc-Program, Else Kröner-Fresenius Foundation.

Functional MRI vs. navigated TMS to optimize M1 seed volume delineation for DTI tractography. A prospective study in patients with brain tumours adjacent to the corticospinal tract.

BACKGROUND: DTI-based tractography is an increasingly important tool for planning brain surgery in patients suffering from brain tumours. However, there is an ongoing debate which tracking approaches yield the most valid results. Especially the use of functional localizer data such as navigated transcranial magnetic stimulation (nTMS) or functional magnetic resonance imaging (fMRI) seem to improve fibre tracking data in conditions where anatomical landmarks are less informative due to tumour-induced distortions of the gyral anatomy. We here compared which of the two localizer techniques yields more plausible results with respect to mapping different functional portions of the corticospinal tract (CST) in brain tumour patients. METHODS: The CSTs of 18 patients with intracranial tumours in the vicinity of the primary motor area (M1) were investigated by means of deterministic DTI. The core zone of the tumour-adjacent hand, foot and/or tongue M1 representation served as cortical regions of interest (ROIs). M1 core zones were defined by both the nTMS hot-spots and the fMRI local activation maxima. In addition, for all patients, a subcortical ROI at the level of the inferior anterior pons was implemented into the tracking algorithm in order to improve the anatomical specificity of CST reconstructions. As intra-individual control, we additionally tracked the CST of the hand motor region of the unaffected, i.e., non-lesional hemisphere, again comparing fMRI and nTMS M1 seeds. The plausibility of the fMRI-ROI- vs. nTMS-ROI-based fibre trajectories was assessed by a-priori defined anatomical criteria. Moreover, the anatomical relationship of different fibre courses was compared regarding their distribution in the anterior-posterior direction as well as their location within the posterior limb of the internal capsule (PLIC). RESULTS: Overall, higher plausibility rates were observed for the use of nTMS- as compared to fMRI-defined cortical ROIs (p < 0.05) in tumour vicinity. On the non-lesional hemisphere, however, equally good plausibility rates (100%) were observed for both localizer techniques. fMRI-originated fibres generally followed a more posterior course relative to the nTMS-based tracts (p < 0.01) in both the lesional and non-lesional hemisphere. CONCLUSION: NTMS achieved better tracking results than fMRI in conditions when the cortical tract origin (M1) was located in close vicinity to a brain tumour, probably influencing neurovascular coupling. Hence, especially in situations with altered BOLD signal physiology, nTMS seems to be the method of choice in order to identify seed regions for CST mapping in patients.

Improved nTMS- and DTI-derived CST tractography through anatomical ROI seeding on anterior pontine level compared to internal capsule.

Imaging of the course of the corticospinal tract (CST) by diffusion tensor imaging (DTI) is useful for function-preserving tumour surgery. The integration of functional localizer data into tracking algorithms offers to establish a direct structure-function relationship in DTI data. However, alterations of MRI signals in and adjacent to brain tumours often lead to spurious tracking results. We here compared the impact of subcortical seed regions placed at different positions and the influences of the somatotopic location of the cortical seed and clinical co-factors on fibre tracking plausibility in brain tumour patients. The CST of 32 patients with intracranial tumours was investigated by means of deterministic DTI and neuronavigated transcranial magnetic stimulation (nTMS). The cortical seeds were defined by the nTMS hot spots of the primary motor area (M1) of the hand, the foot and the tongue representation. The CST originating from the contralesional M1 hand area was mapped as intra-individual reference. As subcortical region of interests (ROI), we used the posterior limb of the internal capsule (PLIC) and/or the anterior inferior pontine region (aiP). The plausibility of the fibre trajectories was assessed by a-priori defined anatomical criteria. The following potential co-factors were analysed: Karnofsky Performance Scale (KPS), resting motor threshold (RMT), T1-CE tumour volume, T2 oedema volume, presence of oedema within the PLIC, the fractional anisotropy threshold (FAT) to elicit a minimum amount of fibres and the minimal fibre length. The results showed a higher proportion of plausible fibre tracts for the aiP-ROI compared to the PLIC-ROI. Low FAT values and the presence of peritumoural oedema within the PLIC led to less plausible fibre tracking results. Most plausible results were obtained when the FAT ranged above a cut-off of 0.105. In addition, there was a strong effect of somatotopic location of the seed ROI; best plausibility was obtained for the contralateral hand CST (100%), followed by the ipsilesional hand CST (>95%), the ipsilesional foot (>85%) and tongue (>75%) CST. In summary, we found that the aiP-ROI yielded better tracking results compared to the IC-ROI when using deterministic CST tractography in brain tumour patients, especially when the M1 hand area was tracked. In case of FAT values lower than 0.10, the result of the respective CST tractography should be interpreted with caution with respect to spurious tracking results. Moreover, the presence of oedema within the internal capsule should be considered a negative predictor for plausible CST tracking.