A fuzzy feature fusion method for auto-segmentation of gliomas with multi-modality diffusion and perfusion magnetic resonance images in radiotherapy

Abstract

The diffusion and perfusion magnetic resonance (MR) images can provide functional information about tumour and enable more sensitive detection of the tumour extent. We aimed to develop a fuzzy feature fusion method for auto-segmentation of gliomas in radiotherapy planning using multi-parametric functional MR images including apparent diffusion coefficient (ADC), fractional anisotropy (FA) and relative cerebral blood volume (rCBV). For each functional modality, one histogram-based fuzzy model was created to transform image volume into a fuzzy feature space. Based on the fuzzy fusion result of the three fuzzy feature spaces, regions with high possibility belonging to tumour were generated automatically. The auto-segmentations of tumour in structural MR images were added in final auto-segmented gross tumour volume (GTV). For evaluation, one radiation oncologist delineated GTVs for nine patients with all modalities. Comparisons between manually delineated and auto-segmented GTVs showed that, the mean volume difference was 8.69% (±5.62%); the mean Dice's similarity coefficient (DSC) was 0.88 (±0.02); the mean sensitivity and specificity of auto-segmentation was 0.87 (±0.04) and 0.98 (±0.01) respectively. High accuracy and efficiency can be achieved with the new method, which shows potential of utilizing functional multi-parametric MR images for target definition in precision radiation treatment planning for patients with gliomas.

Figure 1

Multimodal images of a patient with glioblastoma multiforme (GBM). (a) T1-weighted contrast-enhanced image, (b) T2-weighted image, (c) apparent diffusion coefficient (ADC) map, (d) fractional anisotropy (FA) map, (e) relative cerebral blood volume (rCBV) map. The appearances of tumor in anatomical and functional images are different from each other.

Figure 2

Framework of fuzzy fusion and tumor segmentation using multi-modality diffusion and perfusion MRI images. The original input images are noted MRI sequences of T1-weighted contrast-enhanced images (T1C), T2-weighted images (T2), and parameter maps of apparent diffusion coefficient (ADC), fractional anisotropy (FA) and relative cerebral blood volume (rCBV). Final tumor extent is the combination of segmentation results derived from functional and anatomical parts as shown in the left and right columns.

Figure 3

The membership functions of the tumorous tissue in multi-parametric MRI images. (a) Membership function in apparent diffusion coefficient map (ADC) noted as $M{F}_{ADC}({\nu }_{ADC})$, (b) membership function in fractional anisotropy map (FA) noted as $M{F}_{FA}({\nu }_{FA})$, (c) membership function in relative cerebral blood volume map (rCBV) noted as $M{F}_{CBV}({\nu }_{CBV})$.

Figure 4

One slice of three parametric MRI image volumes and the corresponding three fuzzy feature spaces, in which voxel values represent the probability belonging to tumour. (a) Fractional anisotropy map (FA), (b) apparent diffusion coefficient map (ADC), (c) relative cerebral blood volume map (rCBV), (d) fuzzy feature space of FA map noted as $F{S}_{FA}$, (e) fuzzy feature space of ADC map noted as $F{S}_{ADC}$, (f) fuzzy feature space of rCBV map noted as $F{S}_{CBV}$.

Figure 5

The fused fuzzy feature volume (a) and regions with different memberships. (b) ≥0.6, (c) ≥0.7, (d) ≥0.8.

Figure 6

Comparison between manually delineated gross tumour volume (GTV) and automatically segmented GTV on axial slices of T2-weighted MRI images for 9 patients. (a–i) Represent patient No. 1~9 respectively.