Background: Colon tumor is a molecularly heterogeneous disease. Consequently, tissue sampling biopsy cannot be considered fully representative of tumor behaviour and molecular profile. Such limitations have led researchers to seek alternative procedures able to comprehensively assess the whole tumour heterogeneity. The "Omic" revolution has recently involved also the field of medical imaging, through radiomics. Such technology aims at extracting quantitative features from imaging studies to improve disease characterization. The possibility to merge radiomic data derived from the whole tumour with genomics, might allow to create a model that can be used to characterize colon tumors and predict their behaviors in terms of aggressiveness, relapse and risk of metastases.
Aims: The ADRENALIN (rADiogenomic in ColoREctal Cancer: aN ArtificiaL IntelligeNce approach) project will be a 2-year project with the main aim to characterize biological processes in a voxel-wise high-spatial resolution approach, extracting first, second and third order radiomics features from CT datasets of patients with colorectal tumor. Radiomic features will be subsequently computed by AI- model and integrated with clinical data to eventually generate a radiogenomic signature for personalised management of patients with colon tumor.
Experimental Design: A retrospective data sets collection of 100 patients (radiology, pathology, genomics) will be analysed in order to identify the genetic and radiomic features of colorectal tumors and the clinical endpoints as the outcomes of the predictive model. Further, a prospective population of 50 patients will be enrolled to obtain a new multicentric population for the test and validation of the AI-model.
Expected Results: AI-model including different cancer hallmarks (radiology, pathology, genomic) may represent a comprehensive assessment of colorectal tumor with increased
accuracy in terms of response to treatment and prognosis.
The ADRENALIN (rADiogenomic in ColoREctal Cancer: aN ArtificiaL IntelligeNce approach) project, leading to the development of a radiogenomic signature of colorectal tumor, will have a great impact in oncology, being beneficial for both patients and medical professionals. A strong impact will be found at different levels, from research to patient care, thus involving not only high-end oncologic research centres, but also oncologic multidisciplinary teams working in community hospitals. The development of a radiogenomic signature will overcome the main limitation of conventional biopsies, incomplete tissue sampling, which might lead to a misinterpretation of real tumor characteristics and consequently to a possible undertreatment. Further advantage will be to improve safety of oncologic Patients, in particular those with co-morbidities, avoiding useless and risky biopsies.
Artificial intelligence (AI) algorithms, particularly deep learning, have demonstrated remarkable progress in image-recognition tasks. AI methods excel at automatically recognizing complex patterns in imaging data and providing quantitative, rather than qualitative, assessments of radiographic characteristics.
Preliminary data regarding AI and gastrointestinal oncologic studies were applied using a novel method to automatically segment rectal cancer from 3D MR images based on combination of 3D fully convolutional neural networks (3D- FCNNs) and 3D level-set, demonstrating that the proposed method gives better and accurate segmentation results than 3D-FCNNs alone. Recently, a novel deep learning-based algorithm suited for volumetric colorectal tumors segmentation has been proposed to automatically segment colorectal tumors in 3D T2-weighted MRI with reasonable accuracy. The proposed CNN architecture, based on densely connected neural network, contains multiscale dense interconnectivity between layers of fine and coarse scales, thus leveraging multiscale contextual information in the network to get better flow of information throughout the network. In addition, the 3D level set algorithm was included as a postprocessing task to refine contours of the network predicted segmentation. State of the art methods proposed for tissue segmentation were CT, used for the majority of the methods, and a neural network approach as the second most commonly implemented algorithm after atlas-based methods. Moreover, Deep learning techniques and in particular CNN and fCNN were found to be the most promising and innovative strategies for supervised segmentation task. In a recent study about rectal cancer, Haralick features derived from Texture analysis, a novel oncologic imaging biomarker, were used to assess quantitatively the heterogeneity within a tumor in predicting tumor response to neoadjuvant chemoradiotherapy (CRT). Patients with locally advanced rectal cancer who underwent pre-treatment MRI were enrolled and divided into two groups based on histological response to neoadjuvant CRT in complete responders (CR) and non-responders (NR). All the examinations before CRT were analyzed by two radiologists in consensus and 14 over 192 Haralick¿s features were extrapolated from normalized gray-level co-occurrence matrix in four different directions. Results showed that pretreatment MRI examination showed significant value (p
This artificial intelligence-based model approach will be based on the extraction of radiomic quantitative parameters from Computed Tomography images of colon tumors (Stage II-III) combined with pathological, genomic and transcriptomic information. This model will allow for more accurate and precise understanding of tumor behavior, aggressiveness, and prognosis, ultimately impacting the therapeutic strategies.