Radiomic Features from Post-Operative 18F-FDG PET/CT and CT Imaging Associated with Locally Recurrent Rectal Cancer: Preliminary Findings

Abstract

Locally Recurrent Rectal Cancer (LRRC) remains a major clinical concern; it rapidly invades pelvic organs and nerve roots, causing severe symptoms. Curative-intent salvage therapy offers the only potential for cure but it has a higher chance of success when LRRC is diagnosed at an early stage. Imaging diagnosis of LRRC is very challenging due to fibrosis and inflammatory pelvic tissue, which can mislead even the most expert reader. This study exploited a radiomic analysis to enrich, through quantitative features, the characterization of tissue properties, thus favoring an accurate detection of LRRC by Computed Tomography (CT) and 18F-FDG-Positron Emission Tomography/CT (PET/CT). Of 563 eligible patients undergoing radical resection (R0) of primary RC, 57 patients with suspected LRRC were included, 33 of which were histologically confirmed. After manually segmenting suspected LRRC in CT and PET/CT, 144 Radiomic Features (RFs) were generated, and RFs were investigated for univariate significant discriminations (Wilcoxon rank-sum test, p < 0.050) of LRRC from NO LRRC. Five RFs in PET/CT (p < 0.017) and two in CT (p < 0.022) enabled, individually, a clear distinction of the groups, and one RF was shared by PET/CT and CT. As well as confirming the potential role of radiomics to advance LRRC diagnosis, the aforementioned shared RF describes LRRC as tissues having high local inhomogeneity due to the evolving tissue’s properties.

An Apparent Diffusion Coefficient-Based Machine Learning Model Can Improve Prostate Cancer Detection in the Grey Area of the Prostate Imaging Reporting and Data System Category 3: A Single-Centre Experience

“…For instance, two studies [ 21 , 22 , 23 , 24 , 25 ] report exclusively AUC values. As regards the study in [ 20 ], not any validation, either internal or external, is reported, so the AUC value should be compared to that achieved in the training subset, which is somewhat higher. Moreover, results in [ 21 ] are achieved with a notably high number of RFs with respect to their sample size.…”

“…From a radiological point of view, despite the improvements introduced by subsequent versions of PI-RADS, it is now clear that it is not possible to overcome intrinsic limitations of the visual analysis of the images. Radiomics have become popular, as the employment of well-established machine learning and artificial intelligence strategies for analysing radiological images can enrich the information retrieved by visual analysis, even, in fact, beyond that perceivable by the human eye, by means of the generation of quantitative measurements, called radiomic features (RFs), and the employment of the most relevant ones to build a predictive model supporting clinicians’ decisions [ 18 , 19 , 20 ]. Potentially, using radiomics in PI-RADS 3 lesions evaluation could allow for the diagnosis of benign and malignant lesions by analysing image features only, definitely improving the PCa predictive capability of the PI-RADS 3 score.…”