Background and AimsInternet and social media platforms have become an unprecedented source for sharing self‐experience, potentially allowing the collection and integration of health data with patient experience. StuffThatWorks (STW) is an online open platform that applies machine learning and the power of crowdsourcing, where patients with chronic medical conditions can self‐report and compare their individual outcomes using a structured online questionnaire. We aimed to conduct a cross‐sectional, international, crowdsourcing, artificial‐intelligence (AI) web‐based study of patients with Crohn's disease (CD) self‐reporting their outcomes.MethodsA proprietary STW Bayesian inference model was built to measure improvement in CD severity (on scale of 1–5) for each treatment and ranked treatments using effectiveness. The effectiveness of first‐line biological treatments was analyzed by multiple comparisons and by calculating odds ratios and 95% confidence intervals for each treatment pair.ResultsWe included 7593 self‐reported CD patients for the analysis. Most of the participants were female (75.8%) and from English‐speaking countries (95.7%). Overall, anti‐TNF drugs were the most reported tried treatment (52.8%). Infliximab (IFX) was ranked as the most effective treatment by the STW effectiveness model followed by bowel surgery (second), adalimumab (ADA, third), ustekinumab (UST, 4rd), and vedolizumab (VDZ, fifth). In paired comparison analyses, IFX was most effective, ADA had similar effectiveness compared to UST and all three were more effective than VDZ.ConclusionWe present the first online crowdsourcing AI platform‐based study of self‐reported treatment effectiveness in CD. Net‐based crowdsourcing patient‐reported outcome platforms can potentially help both clinicians and patients select the best treatment for their condition.

Triple-negative breast cancer (TNBC) is a type of aggressive breast cancer lacking the expression of estrogen receptors (ER), progesterone receptors (PR), and human epidermal growth factor receptor-2 (HER-2). TNBC represents a heterogeneous subtype of breast cancer that carries a poorer prognosis. There remains a need to identify novel driver and modifier genes of TNBC, representing targets to treat the disease. MET tyrosine kinase is the receptor for hepatocyte growth factor/scatter factor (HGF/SF). MET overexpression and constitutively activated mutation promotes tumor growth, cell motility and metastases, making MET an ideal target for anti-metastatic precision therapy. In mice carrying MET constitutive activation mutation (METmut), the gene initiates tumorigenicity associated with inherited driver modifier genes (IDMGs). The Collaborative Cross (CC) animal model consists of recombinant inbred lines generated by reciprocal crosses between numerous founder lines. It is a reference to apply genetic traits mapping with high-resolution phenotypes such as radiomic features. In this study, we established a new machine learning prognostic model of overall survival (OS) of METmut-induced mammary carcinoma. QTL (Quantitative Trait Locus) mapping analysis was used to identify candidate genes that may contribute to intra-tumor heterogenicity and OS. Two CT images cohorts were included: METmut-FVB/N and METmut-CC mice that developed mammary carcinoma (n=13 and n=15, respectively). Each segmented tumor was divided into 3D cubes, and radiomic features were extracted from each cube using a novel in-house tool. We performed Principle Component Analysis (PCA) and cluster analysis (k-means) on the radiomic features to produce cube cluster groups with similar radiomic characteristics, such as tumor region intensity and morphology. We then assessed each cluster group's proportion per tumor – the cluster proportion - as a radiomic phenotype for QTL. QTL analysis identified loci in chromosomes #6 and #15 that code for new radiomics morphological candidate genes that may alter MET signaling and reduce OS. To analyze tumor heterogenicity, median and standard deviation of cubes PCA values were calculated per tumor and used as covariates in Cox regression against OS. Four out of the eight covariates were significantly correlated with OS (P < 0.01, Xi 21.59 on 8 df). Using PCA and cluster analysis in radiomic feature analysis can create a unique prognostic signature for breast cancer patients. CT-radiomics can be used to isolate tumor genes, substituting tumor biopsies in patients with breast cancer metastases to brain and providing novel targets for precision therapy. Citation Format: Ilan Tsarfaty, Eran Dotan, Michal Bloom, Judith Horev, Fuad Iraqi. AI and CT-radiomics as a tool for breast cancer prognosis and radiomics-gene isolation: Activated MET induced mammary carcinoma as a model [abstract]. In: Proceedings of the AACR Virtual Special Conference on Artificial Intelligence, Diagnosis, and Imaging; 2021 Jan 13-14. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(5_Suppl):Abstract nr PO-067.