Purpose Lung cancer tumours treated with radiotherapy may display elastic (true shrinkage) or non-elastic (eroding) changes. Adapting treatment fields to non-elastic shrinkage can lead to a potential treatment failure due to underdosage of residual microscopic disease. Therefore, we propose a new methodology to differentiate between these two modes of regression, utilising on-treatment cone-beam computed tomography (CBCT) images. Our hypothesis is that, at the lung/tumour boundary, erosion will lead to a gradual decrease in Hounsfield units (HU) while true shrinkage will result in faster HU changes, as low-density lung tissue is returning to replace tumour tissue. Methods CBCT data sets from 8 non-small cell lung cancer (NSCLC) patients treated with 55Gy in 20 fractions were included in this study. The tumour volume was automatically thresholded on the day one CBCT. An annular region of interest (ROI) was created from the day one contour (2mm inside and 2mm outside; excluding mediastinium/chest-wall), and this ROI was rigidly propagated across all CBCTs. Histograms of HU within these ROIs were fitted with a bimodal Gaussian distribution. It is assumed that the lower HU peak corresponds to healthy lung tissue and the higher HU peak corresponds to cancerous tissue. The change in the relative heights of the peaks was used to determine the mode of regression of the tumour. Results Visual evaluation of the motion of anatomical landmarks (e.g bronchioles) classified 5 tumours as displaying non-elastic changes (eroding) while 3 displayed elastic changes (shrinking). Shrinking tumours displayed a larger and more consistent change in relative peak heights throughout the treatment course, with changes appearing rapidly. In comparison, eroding tumours showed a more gradual change in relative peak heights. Additionally, eroding tumours showed a lower correlation coefficient than shrinking tumours, i.e. greater variability in relative heights. However, the small patient numbers prevented robust stastical analysis of these differences. Conclusion We believe that change of statistics of the HU in the tumour rim has the potential to differentiate between eroding and shrinking tumours. We will next apply this method in a much larger patient cohort.