"Epithelial ovarian cancer (EOC) is the sixth most common cancer among women in the UK and has the highest mortality of all gynecological cancers, accounting for 4% of all cancer deaths in women. High-grade serous ovarian cancer (HGSOC) represents the most dominant (70% of EOC patients) and most lethal histological subtype. Although it is well known that HGSOC patients have a heterogeneous response to treatment and prognosis, extensive cytoreductive surgery combined with platinumbased chemotherapy are currently the standard treatments for most patients without consideration of individual prognostic and predictive biomarkers".
"Using machine learning, we derive a non-invasive summary-statistic of the primary ovarian tumor based on 4 descriptors, which we name “Radiomic Prognostic Vector” (RPV). The prognostic model we propose is simple, built solely on the information extracted from a patient’s routine preoperative CT scan at the presentation of the disease and hence readily accessible without additional costs or time delays, knowing that majority of the HGSOC patients will have CT scans prior to the treatment (compared to PET, MRI or ultrasound). The entire primary ovarian mass is segmented, signifying that any prognostic or biological information extracted is more representative of the disease compared to a single site biopsy. Moreover, RPV is stable across the CT scanner types and the segmentation process, thus limiting the number of potential restrictions for clinical exploitation in the future. We have constructed a software pipeline which is able to compute the RPV of 80 EOC datasets within 5 min on a standard computer. Beyond RPV, the dataset could be mined in a supervised manner for new gene- or protein-radiomics interactions".
"We wished to investigate the data structure within the radiomic profiles derived from primary tumors of EOC patients in relation to clinical and genetic features. For samples with both radiomics and CNA data, we performed a spectral clustering analysis based on the Pearson correlation coefficients between each samples’ radiomic profile. There was a clear division of samples into three major groups with each group characterized by high feature similarity but largely distinct from those in other groups. In aggregate, unsupervised analysis highlighted an intrinsic association between radiomic profile, genetic background, and clinical characteristics, warranting further characterization. To further understand the radiomic characteristics of the HGSOC subtype, we performed unsupervised hierarchical clustering analysis... unsupervised analysis highlighted an intrinsic association between radiomic profile, genetic background, and clinical characteristics, warranting further characterization. With an unsupervised k-means clustering approach, we split all the patients from the three cohorts based on their RPV into three subgroups (low risk, medium risk, and high risk)".
The researchers "developed TexLab 2.0, a software program that summarized 657 features relating to the shape and size, intensity, texture and wavelet decompositions of 364 preoperative contrast-enhanced CT scans16.. A comprehensive molecular profile including gene expression, copy-number, and protein expression was analyzed for a subset of patients. 294 primary EOC patients with fresh frozen tissue treated within the Hammersmith Hospital, Imperial College Healthcare NHS Trust, London, UK between 2004 and 2015 as well as 70 EOC patients from the TCGA project".
RPV reliably identifies the 5% of patients with median overall survival less than 2 years, significantly improves established prognostic methods, and is validated in two independent, multi-center cohorts. Furthermore, genetic, transcriptomic and proteomic analysis from two independent datasets elucidate that stromal phenotype and DNA damage response pathways are activated in RPV-stratified tumors. RPV and its associated analysis platform could be exploited to guide personalized therapy of EOC and is potentially transferrable to other cancer types.