Automatic Segmentation of Heschl Gyrus and Planum Temporale by MRICloud

Abstract

Objectives: This study used a cloud-based program, MRICloud, which parcellates T1 MRI brain scans using a probabilistic classification based on manually labeled multi-atlas, to create a tool to segment Heschl gyrus (HG) and the planum temporale (PT).

Methods: MRICloud is an online platform that can automatically segment structural MRIs into 287 labeled brain regions. A 31-brain multi-atlas was manually resegmented to include tags for the HG and PT. This modified atlas set with additional manually labeled regions of interest acted as a new multi-atlas set and was uploaded to MRICloud. This new method of automated segmentation of HG and PT was then compared to manual segmentation of HG and PT in MRIs of 10 healthy adults using Dice similarity coefficient (DSC), Hausdorff distance (HD), and intraclass correlation coefficient (ICC).

Results: This multi-atlas set was uploaded to MRICloud for public use. When compared to reference manual segmentations of the HG and PT, there was an average DSC for HG and PT of 0.62 ± 0.07, HD of 8.10 ± 3.47 mm, and an ICC for these regions of 0.83 (0.68-0.91), consistent with an appropriate automatic segmentation accuracy.

Conclusion: This multi-atlas can alleviate the manual segmentation effort and the difficulty in choosing an HG and PT anatomical definition. This protocol is limited by the morphology of the MRI scans needed to make the MRICloud atlas set. Future work will apply this multi-atlas to observe MRI changes in hearing-associated disorders.

Keywords: Hearing Loss; Otology/Neurotology; Radiology.

FIG. 1.

Representative images for manual vs MRICloud segmentation of HG and PT. Subject had a complete posterior duplication (CPD) on the right side and a single HG on the left side. T1 image with both manual and MRICloud segmentation represented in axial, coronal, and sagittal views. Dice scores for overlap are the following: left HG = 0.747, right HG = 0.678, left PT = 0.610, right PT = 0.642. HG indicates Heschl gyrus; PT, planum temporale.

FIG. 2.

Size dependency of segmentation accuracy. There was no relationship between the DSC of the HG and PT and their reference segmentation volume: A, HG (y = 0.00003x + 0.588, R² = 0.060, P = 0.298). B, PT (y = 0.00008x + 0.443, R² = 0.191, P = 0.054). DSC indicates dice similarity coefficient; HG, Heschl gyrus; PT, planum temporale.

FIG. 3.

Size estimations of HG and PT affect segmentation quality. There was a strong correlation between the MRICloud segmentation volume and reference segmentation volume for HG but not for the PT: A, HG (y = 0.598x + 549, R² = 0.675, P < 0.001). B, PT (y = 0.279x + 1367, R² = 0.184, P = 0.059). The volume ratio of the MRICloud and reference segmentation predicted the segmentation accuracy for the PT but not the HG. C, HG (y = −0.091x + 0.724, R² = 0.055, P = 0.319). D, PT (y = −0.219x + 0.828, R² = 0.318, P = 0.010). The DSC was strongly correlated for the right HG and PT but not the left HG and PT. E, Left (y = 0.319x + 0.454, R² = 0.233, P = 0.158). F, Right (y = 0.504x + 0.319, R² = 0.597, P = 0.009). DSC indicates dice similarity coefficient; HG, Heschl gyrus; PT, planum temporale.

FIG. 4.

Segmentation images for 2 separate subjects with T1 image and reference segmentation in axial (A,D), coronal (B,E), and sagittal (C,F) views. The dark blue is the right PT and the light blue is right HG. On top is a good segmentation of HG and PT (A,B,C) with a DSC of 0.70 for HG and 0.76 for PT. The bottom images are an example of poor segmentation with a DSC of 0.53 for HG and 0.52 for PT. DSC indicates dice similarity coefficient; HG, Heschl gyrus; PT, planum temporale.