Skip to content

New Tool Measures Glioma Response to Treatment Automatically

Key findings

  • AutoRANO algorithm was developed to measure tumor burden according to the Response Assessment in Neuro-Oncology (RANO) criteria
  • In this study the algorithm was validated in preoperative and postoperative glioma patient cohorts
  • Automated tumor volume and RANO measurements matched human expert ratings in terms of change over the course of treatment
  • The new tool may expedite measurement of tumor burden for evaluation of treatment response, decrease clinician burden associated with manual tumor segmentation and decrease interobserver variability

To assess treatment response in patients with glioma, the gold standard is to monitor changes in tumor size using magnetic resonance imaging (MRI). For high-grade gliomas, this requires:

  • Measuring the two-dimensional product of the maximum bidimensional diameters of contrast-enhancing tumor
  • Qualitatively evaluating the image for abnormal regions of fluid-attenuated inversion recovery (FLAIR) hyperintensity

This procedure is not only time-consuming but also subject to interobserver variability. As previously noted in Neuro-Oncology, volumetric measurements may capture tumor burden more accurately, especially since gliomas are often irregularly shaped. However, they too are labor-intensive because of the effort needed to perform tumor segmentation (determine the tumor's boundaries).

Jayashree Kalpathy-Cramer, PhD, of the Athinoula A. Martinos Center for Biomedical Imaging at Massachusetts General Hospital, and colleagues now report in Neuro-Oncology on their new open-source software, called AutoRANO, for automated volumetric and bidimensional measurement of glioma burden.


AutoRANO, automatic volume according to the Response Assessment in Neuro-Oncology (RANO) criteria, is a deep learning algorithm developed for automatic segmentation of medical images, now that more powerful graphics processing programs are available. At the core of deep learning is the convolutional neural network, a machine learning technique that can be trained on raw image data to predict clinical outputs of interest.

In AutoRANO, all of the following operations are fully automated:

  • "Brain extraction," a pre-processing technique that prevents the algorithm from falsely labeling the skull as brain
  • Segmentation of abnormal FLAIR hyperintensity and contrast-enhancing tumor
  • Quantification of tumor volume and the product of maximum bidimensional diameters according to the RANO criteria

Validation Cohorts

The researchers validated AutoRANO in two cohorts:

  • Preoperative: 843 patients at four institutions who had MRI performed before surgery for grade II to IV glioma. The scans consisted of T2-weighted FLAIR and post-contrast T1-weighted images
  • Postoperative: 713 MRIs from 54 patients at Mass General who had newly diagnosed glioblastoma. They were part of a trial of chemoradiation or a trial of chemoradiation with cediranib after surgery. MRI, including FLAIR and both pre- and post-contrast T1-weighted images, were obtained at various points during treatment

Manual vs. AutoRANO Assessments


RANO Measurements

On average, the RANO measurements from AutoRANO were larger than those of two human expert raters. Moreover, the average RANO measurements differed between the two raters. This is probably because AutoRANO performs an exhaustive search of the longest perpendicular diameters while a human can only estimate by eye, a less accurate method.

Contrast-enhancing Tumor and FLAIR Hyperintensity

There was high agreement between human raters and automatic volume for both contrast-enhancing tumor and FLAIR hyperintensity. This suggests that volume measurements allow for greater consistency across raters than RANO measurements do.

Tumor Burden

Over time in the postoperative cohort, there was high agreement between human raters and AutoRANO with regard to changes in tumor burden (both contrast-enhancing and FLAIR hyperintensity).

Interestingly, AutoRANO correlated better with manual contrast-enhancing tumor volume than the manual RANO measurements. Similarly, change in AutoRANO over time correlated better with the change in manual contrast-enhancing tumor volume than the change in manual RANO measurements.

This suggests AutoRANO may be a more accurate measure of tumor burden than manual RANO measurement is, in addition to the advantage of being fully automated.

AutoRANO was successful even in the patients who were treated with cediranib, which blunts contrast enhancement, yielding ill-defined contrast enhancement margins that are difficult to contour. It is in these cases, particularly, that automated tumor segmentation is likely to be quite helpful.

Future Applications

This study serves as proof-of-concept for the application of AutoRANO to other tumor types. AutoRANO is expected to expedite measurement of tumor burden for evaluation of treatment response, decrease clinician burden associated with manual tumor segmentation and decrease interobserver variability.

AutoRANO might also facilitate the use of tumor volume as a response endpoint in clinical trials.

Learn more about the Athinoula A. Martinos Center for Biomedical Imaging

Refer a patient to the Department of Neurosurgery


A phase II trial kicked off at Massachusetts General Hospital involving multiple departments and research institutes are utilizing oncolytic viruses to treat glioblastomas.


Working with colleagues in the Netherlands, a neuroscientist at Massachusetts General Hospital has demonstrated for the first time that noninvasive measurement of brain activity with magnetoencephalography can predict progression-free survival in patients with glioma and might even lead to new therapeutic strategies.