MRI is the imaging modality of choice for diagnosis and follow-up of soft-tissue sarcomas. T₁- and T₂-weighted images allow tumor detection and characterization. Diffusion weighted MRI (DW-MRI) is a useful addition to conventional MRI for its ability to probe the tumor microenvironment. Pathological tissue signatures within the tumor, such as high cellularity, high T₂ content, or necrosis, can be interpreted from the combination of T₂-weighted images, DW-MRI (b=500-1000 s/mm²) and ADC maps [1]. Delineating tumor sub-regions is useful in assessing the tumor microenvironment and potentially provides information about sub-region targets for radiation dose-painting. Conventionally, these tissue types are interpreted by visual inspection, which can be time-consuming and subjective. In this work, we propose a novel method to automatically distinguish various pathological tissue types within a tumor, using normal muscle tissue as a reference. To automate the tissue segmentation process, we identified quantities that reflect the main physical characteristics captured by T₂-weighted and DW-MRI. The measured T₂ value is the physical characteristic reflected in T₂-weighted images. The signal for DW-MRI with a non-zero b-value can be expressed as $$$S_{(b,TE)SE}=S_0\cdot exp(-\frac{TE}{T_2})\cdot exp(-b\cdot ADC)$$$. The product of these two exponentials is used to construct a surrogate map representing the high b-value DW-MRI, or simulated DWI (simDWI), which reflects the relative signal attenuation due to T₂ and diffusion. In the resulting segmentation approach, we replace the T₂-weighted MRI and DW-MRI with the T₂ map and simDWI respectively. The trio of values (T₂, simDWI, ADC) in each tumor voxel is compared against the average values from a reference tissue (i.e. muscle). Each tumor voxel is assigned to one of four distinct pathological classes according to the parameter values relative to the reference, demonstrated in Table 1. To visualize this segmentation, each class of tissue is represented by a different color. The degree of confidence with which a voxel belongs to a given class is quantified by $$$\sqrt{\sum\limits_{k=1}^3 (\frac{x_k-\mu_{m,k}}{\sigma_{m,k}})^2}$$$, where x repre­­sents the value of the parameter in the voxel, the subscripts k =1, 2, 3 denote the T₂, simDWI, and ADC respectively, $$$\mu_m$$$ and $$$\sigma_m$$$ denote the global mean and the standard deviation of reference tissue parameters. In the visualization, confidence is assigned to the color saturation.Axial 2D DW-MRI and 2D fast spin echo (FSE) images with fat saturation were acquired on a 1.5 T scanner (GE Healthcare) in 5 patients with biopsy-confirmed soft-tissue sarcoma. DW-MRI were acquired with b = 100, 800 s/mm², TE = 88 ms and TR = 5000 ms. For FSE images, the TR ranged between 3.95 s and 6.65 s; the echo train length (ETL) was 9. FSE data were acquired twice – for T₂ mapping – using a short TE of 9 to 12 ms and a long TE of 64 to 83 ms. The FOV, number of slices, and slice thickness were adapted for each patient. ADC and T₂ values were computed for each image voxel using MATLAB (The Mathworks Inc). T₂ maps were registered and resampled to the DW-MRI space using MIMVista (MIM Software Inc). Muscle regions were identified on the DW-MRI (b = 0 s/mm²) for each patient. The reference mean and standard deviation ADC, T₂, and simDWI values were calculated including muscle voxels from all patients in the study.

Based on the collective intensity patterns of the quantitative T2, simDWI, and ADC maps, regions of high T2 content, high cellularity tumor, and necrosis were distinguished for various tumor types (Figure 1). A representative case featuring the segmentation of myxoid/round cell liposarcoma is shown in Figure 2. The segmentation results indicate that the lesion is composed of 90% high T2 content (red) and 2% high cellularity content (green), which are consistent with histopathological observations from biopsy. The overall histology shows large area of low-grade myxoid matrix (high T2), with < 5% of high grade round cells. A key insight in this method is to represent the high-b-value DW-MRI with the simDWI map. The simDWI carries the relative contribution from T2 and ADC, and play a crucial role in differentiating necrotic tissue from high T2 content. Moreover, since T2 and ADC reflect the intrinsic properties of a given tissue, a global mean muscle T2 and ADC, obtained from a group of patients, can be applied to images without identifiable muscle and to future patients.

We have successfully automated the process of differentiating pathologically important tumor regions including high cellularity, high T₂ content, necrosis, and fibrous tissues. This technique yields useful knowledge about tumor composition and could potentially provide information about sub-region targets for radiation dose-painting.