Soft tissue sarcomas are highly heterogeneous tumours with variable components, which can include cellular tumour, fat, necrosis and cystic change. Post-treatment changes often cannot be described by standard size criteria e.g. RECIST 1.1, as responding tumours may not change size, or may grow, after radiotherapy.^1,2 Furthermore, response assessments for liposarcomas, the commonest sarcoma sub-type, present their own challenges. Liposarcomas consist of varying proportions of well-differentiated (fatty) and dedifferentiated (cellular) components and it is the dedifferentiated component which will respond to treatment. As the two components are often mixed, accurate dimension-based response assessments are extremely challenging.

In non-resectable disease, and in trials of non-surgical treatments, functional imaging may provide non-invasive methods for response assessment. However, evaluation of functional imaging parameters over the whole tumour volume may not reveal the full extent of post-treatment changes, which may occur in localised regions or be confined to one class of tumour components. Segmentation of tumours into regions of variable biological behaviour is also critical for development of radiotherapy protocols where dose is escalated to the most resistant or aggressive components of the tumour.

Methods are required, therefore, to segment highly heterogeneous tumours; evaluate functional imaging parameters over selected components; and provide a visual summary of post-treatment changes by combining multiple functional imaging parameters.

To develop methods for analysis of multi-parametric functional imaging in heterogeneous retroperitoneal sarcomas and assess post-radiotherapy changes.

Patients: Patients with retroperitoneal sarcoma were imaged prior to treatment, with their written consent, as part of a prospective single-centre study. Tumours included leiomyosarcomas, well-differentiated/dedifferentiated liposarcomas and spindle cell sarcomas. In patients receiving pre-operative radiotherapy (50.4Gy in 28 fractions) another MR examination was carried out 2-4 weeks after radiotherapy, prior to surgery.

Imaging: Scans were carried out using a 1.5T MR scanner (Aera, Siemens GmbH, Erlangen, Germany). Axial T₂-w images, diffusion-weighted images (DWI; b-values 50,600,900smm^-2), Dixon and T₁-weighted images (3D FLASH, 17°) were acquired from the whole tumour volume. 4-minutes after administration of Gd-based contrast agent (Dotarem, 0.2ml/kg body-weight, 2ml/sec), post-contrast T₁-w images were acquired for evaluation of enhancing fraction (EF).

Analysis: Regions of interest (ROIs) were drawn around the whole tumour on all slices on which the tumour appeared on T₂-w images by a consultant radiologist. ROIs were applied to apparent diffusion coefficient (ADC) maps, fat fraction (FF) maps, calculated from Dixon, and EF maps, calculated from pre- and post-contrast T₁-w images$$$\;\left(\mathrm{EF}=\left(\mathrm{S}_{post}-\mathrm{S}_{pre}\right)/\left(\mathrm{S}_{post}+\mathrm{S}_{pre}\right)\right)$$$.

For each patient,$$$\;N\;$$$series of parametric images ($$$N$$$=1: ADC, EF or FF;$$$\;N$$$=2: ADC/EF, ADC/FF or EF/FF;$$$\;N$$$=3: ADC/EF/FF) were selected and$$$\;N$$$-dimensional scatter-plots of pixel values in the tumour-volume created.$$$\;N\;$$$was chosen by inspection of the images to determine the parameters characterising the tumour and the same number used pre- and post-radiotherapy. J-class Gaussian mixture modelling was applied, using the expectation maximisation algorithm, to estimate the posterior probability$$$\;p_{ij}\;$$$of the$$$\;i^{th}\;$$$pixel belonging to the class$$$\;\mathrm{C}_{j}$$$; the number and initial position of classes was seeded by the user.^3,4 Resulting maximum a posteriori classification of tissues was displayed as colour-coded regions superimposed on greyscale images.

Examples of 4 patients are shown to illustrate pixel classification and changes post-radiotherapy. Figures 1,2 and 3 demonstrate classification of tumour components based on$$$\;N$$$=1,2 or 3 parameters respectively. Figure 4 shows one tumour exhibiting heterogeneous changes post-radiotherapy$$$\;$$$(RT), with an increase in the high ADC/non-enhancing and low ADC/non-enhancing regions (red and light blue) in the anterior part of tumour; the posterior part of the tumour remained relatively unchanged. Figure 5 shows a tumour exhibiting little or no change post-radiotherapy.

By separating highly heterogeneous tumours into multiple classes, properties of each class can be estimated separately, for example to assess the mean ADC of restricted or water-based component(s).

By combining multiple functional imaging parameters, tissues can be classified according to their properties in all parameters simultaneously. The component of the tumour exhibiting restricted diffusion and high EF might be hypothesised to correspond to viable tumour, indicating that assessment of combined parameters might have prognostic value not obtained from morphological imaging or ADC alone.

The absence of post-radiotherapy changes in Figure 5 is characteristic of well-differentiated liposarcomas, which do not respond to radiotherapy unlike the dedifferentiated disease in Figure 4. Although both well- and dedifferentiated disease display heterogeneous ADC, there was notable absence of enhancement in the well-differentiated disease, which did not show changes post-radiotherapy.

Separation of heterogeneous retroperitoneal sarcomas into multiple tissue classes allows estimation of volumes of each component and assessment of mean ADCs and EF of each class. Assessment of each class separately allows more detailed investigation of post-treatment changes and combining functional imaging parameters may provide greater insight into tumour behaviour.