The INCISIVE pilots will focus on the following types of cancer: lung, colorectal, breast and prostate. They will be carried out in 8 pilot sites in Greece, Italy, Spain, Cyprus and Serbia, following all applicable ethical procedures, along with privacy and security protocols.
All pilots will be deployed in two phases. In the first phase, an observational study will be performed where the INCISIVE processing pipelines will be used retrospectively to evaluate the proposed system performance, added value and identify novel, potentially valuable prognostic markers. In the second phase, a short interventional study with a small number of participants will allow the use of the INCISIVE AI tools in practice, thus providing additional summative evaluation data for the developed techniques
Accurate classification and segmentation of non-small lung cancer (NSCLC) lesions by utilizing X-ray images aiming to retrieve initial but still essential diagnostic information, as well as CT scans and PET/CT scans for further diagnostic analysis. Explainable AI (XAI) tools will be adopted in an effort to understand which areas of the image contributed to the final classification decision.
Binary classification healthy/non-healthy of the input MRI scans constitutes the initial stage of the Colorectal cancer diagnostic pipeline. Accordingly, the accurate segmentation of ROIs (Regions Of Interest) and the classification as malignant or benign is another major aspect of this pilot study. Additionally, the prediction of metastatic risk and the evaluation of treatment response would be of great interest for the purposes of this AI service. XAI features will be part of this study as well.
Utilization of low-cost imaging modalities like mammographies, aiming to provide reliable and accurate diagnostic services that focus on identification of lesions and calcifications, as well as prediction of BIRADS score and ACR density. MRI scans will also be evaluated as a secondary step and the findings will be combined with clinical data to address elaborate clinical questions. The diagnostic models will also be enriched with explainability features.
MRI slices will be the input to the AI service of this pilot study. An initial classifier decides if the given image contains any kind of malignancy, while a segmentation algorithm extracts the corresponding Regions of Interest. The PIRADS score will be calculated from the extracted segments and an additional classifier will show the probability of recurrence, taking into consideration specific risk factors of the patient. We will also explore the possibilities of predicting the biopsy score directly from the initial MRI scan, allowing the patient to avoid an invasive screening procedure.