Personalised treatment of cancer patients starts with an extensive evaluation of the cancer that should be treated. Public attention is drawn to next generation sequencing as the ultimate selection tool for treatment. However, fortunately most patients with colorectal cancer (CRC) do not need targeted therapies. The introduction of CRC population screening in The Netherlands, together with improvements in local treatment prevent the development of metastatic disease in the majority of patients.
The next challenges are prevention of overtreatment of patients with early CRC and/or with an intermediate risk CRC and adequate treatment of patient with increased risk CRC. The initial treatment decisions for these patients are partly based on histological parameters, such as invasion depth, differentiation grade, lymphatic invasion and extramural vascular invasion. Evidence from systematic reviews, however, suggests that tumour budding (i.e. the presence of single tumour cells or small tumour cell clusters up to four cells at the invasive front of the cancer) is one of the most powerful histological biomarkers that might help making treatment decisions. However, the implementation of this biomarker is hampered by the lack of standardisation in the histological assessment. Digital image analysis has been proposed as a highly reliable method to determine tumour budding, which will facilitate treatment decisions.
In this project we facilitate convolutional neural network in combination with differently stained whole-slide images of CRC to create a algorithm that is capable of detection tumor budding and tumor clusters (PDC). This will form the basis for detailed and reproducible analysis of the degree of budding within and at the invasive margin of CRC (intratumoral vs peritumoral). Location specific quantitative data will be derived for the number of buds and PDC, by calculating these data for every candidate location in a WSI.