Overcoming data scarcity in radiomics/radiogenomics using synthetic radiomic features.

PURPOSE: To evaluate the potential of synthetic radiomic data generation in addressing data scarcity in radiomics/radiogenomics models. METHODS: This study was conducted on a retrospectively collected cohort of 386 colorectal cancer patients (n = 2570 lesions) for whom matched contrast-enhanced CT images and gene TP53 mutational status were available. The full cohort data was divided into a training cohort (n = 2055 lesions) and an independent and fixed test set (n = 515 lesions). Differently sized training sets were subsampled from the training cohort to measure the impact of sample size on model performance and assess the added value of synthetic radiomic augmentation at different sizes. Five different tabular synthetic data generation models were used to generate synthetic radiomic data based on "real-world" radiomics data extracted from this cohort. The quality and reproducibility of the generated synthetic radiomic data were assessed. Synthetic radiomics were then combined with "real-world" radiomic training data to evaluate their impact on the predictive model's performance. RESULTS: A prediction model was generated using only "real-world" radiomic data, revealing the impact of data scarcity in this particular data set through a lack of predictive performance at low training sample numbers (n = 200, 400, 1000 lesions with average AUC = 0.52, 0.53, and 0.56 respectively, compared to 0.64 when using 2055 training lesions). Synthetic tabular data generation models created reproducible synthetic radiomic data with properties highly similar to "real-world" data (for n = 1000 lesions, average Chi-square = 0.932, average basic statistical correlation = 0.844). The integration of synthetic radiomic data consistently enhanced the performance of predictive models trained with small sample size sets (AUC enhanced by 9.6%, 11.3%, and 16.7% for models trained on n_samples = 200, 400, and 1000 lesions, respectively). In contrast, synthetic data generated from randomised/noisy radiomic data failed to enhance predictive performance underlining the requirement of true signal data to do so. CONCLUSION: Synthetic radiomic data, when combined with real radiomics, could enhance the performance of predictive models. Tabular synthetic data generation might help to overcome limitations in medical AI stemming from data scarcity.

Artificial Intelligence-based Quantification of Pleural Plaque Volume and Association With Lung Function in Asbestos-exposed Patients.

PURPOSE: Pleural plaques (PPs) are morphologic manifestations of long-term asbestos exposure. The relationship between PP and lung function is not well understood, whereas the time-consuming nature of PP delineation to obtain volume impedes research. To automate the laborious task of delineation, we aimed to develop automatic artificial intelligence (AI)-driven segmentation of PP. Moreover, we aimed to explore the relationship between pleural plaque volume (PPV) and pulmonary function tests. MATERIALS AND METHODS: Radiologists manually delineated PPs retrospectively in computed tomography (CT) images of patients with occupational exposure to asbestos (May 2014 to November 2019). We trained an AI model with a no-new-UNet architecture. The Dice Similarity Coefficient quantified the overlap between AI and radiologists. The Spearman correlation coefficient ( r ) was used for the correlation between PPV and pulmonary function test metrics. When recorded, these were vital capacity (VC), forced vital capacity (FVC), and diffusing capacity for carbon monoxide (DLCO). RESULTS: We trained the AI system on 422 CT scans in 5 folds, each time with a different fold (n = 84 to 85) as a test set. On these independent test sets combined, the correlation between the predicted volumes and the ground truth was r = 0.90, and the median overlap was 0.71 Dice Similarity Coefficient. We found weak to moderate correlations with PPV for VC (n = 80, r = -0.40) and FVC (n = 82, r = -0.38), but no correlation for DLCO (n = 84, r = -0.09). When the cohort was split on the median PPV, we observed statistically significantly lower VC ( P = 0.001) and FVC ( P = 0.04) values for the higher PPV patients, but not for DLCO ( P = 0.19). CONCLUSION: We successfully developed an AI algorithm to automatically segment PP in CT images to enable fast volume extraction. Moreover, we have observed that PPV is associated with loss in VC and FVC.

Colon cancer CT staging according to mismatch repair status: Comparison and suggestion of imaging features for high-risk colon cancer.

BACKGROUND: Neoadjuvant treatment with either chemotherapy or immunotherapy is gaining momentum in colon cancers (CC). To reduce over-treatment, increasing staging accuracy using computed tomography (CT) is of high importance. PURPOSE: To assess and compare CT imaging features of CC between mismatch repair-proficient (pMMR) and MMR-deficient (dMMR) tumours and identify CT features that can distinguish high-risk (pT3-4, N+) CC according to MMR status. METHODS: Primary staging CTs of 266 patients who underwent primary surgical resection of a colon tumour were retrospectively and independently evaluated by two radiologists. Logistic regression analysis was performed to identify significant associations between imaging features and positive lymph node status. Receiver operating characteristic (ROC) curves of significantly associated features were assessed and validated in an external cohort of 104 patients. RESULTS: Among pT3 tumours only, dMMR CC were significantly larger than pMMR CC in both length and thickness (length 59.39 ± 26.28 mm versus 48.70 ± 23.72, respectively, p = 0.031; thickness 20.54 mm ± 11.17 versus 16.34 ± 8.73, respectively, p = 0.027). For pMMR tumours, nodal internal heterogeneity on CT was significantly associated with a positive lymph node status (odds ratio (OR) = 2.66, p = 0.027), while for dMMR tumours, the largest short diameter of the nodes was associated with lymph node status (OR = 2.01, p = 0.049). The best cut-off value of the largest short diameter of involved nodes was 10.4 mm for dMMR and 7.95 mm for pMMR. In the external validation cohort, AUCs for predicting involved nodes based on the largest short diameter was 0.764 for dMMR tumours using 10 mm size cut-off and 0.624 for pMMR tumours using 7 mm cut-off. CONCLUSION: These data show that CT imaging features of primary CC differ between dMMR and pMMR tumours, suggesting that the assessment of CT-based CC staging should take MMR status into consideration, especially for lymph node status, and thus may help in selecting patients for neoadjuvant treatment.

Identifying high-risk colon cancer on CT an a radiomics signature improve radiologist's performance for T staging?

PURPOSE: To assess the role of radiomics in detection of high-risk (pT3-4) colon cancer and develop a combined model that combines both radiomics and CT staging of colon cancer. METHODS: We included 292 colon cancer patients who underwent pre-operative CT and primary surgical resection within 2 months. Three-dimensional segmentations and CT staging of primary colon tumors were done. From each 3D segmentation of colon tumor, radiomic features were automatically extracted. Logistic regression analysis was performed to identify associations between radiomic features and high-risk (pT3-4) colon tumors. A combined model that integrated both radiomics and CT staging was developed and their diagnostic performance was compared with that of conventional CT staging. Tenfold cross-validation was used to validate the performance of the model and CT staging. RESULTS: The model that combined radiomic features and CT staging demonstrated a significantly better performance in detection of high-risk colon tumors in training set (AUC = 0.799, 95% CI: 0.720-0.839 for combined model and AUC = 0.697, 95% CI = 0.538-0.756 for CT staging only, p < 0.001 for difference). Cross-validation results also demonstrated significantly better detection performance of combined model (AUC = 0.727, 95% Confidence Interval (CI): 0.621-0.777 for combined model and AUC = 0.628, 95% CI = 0.558-0.689 for CT staging only, Boot CI = 0.099). CONCLUSION: CT radiomic features of primary colon cancer, combined with CT staging, can improve the detection of high-risk colon cancer patients.

Rectal cancer response to neoadjuvant chemoradiotherapy evaluated with MRI: Development and validation of a classification algorithm.

OBJECTIVE: The aim of this study was to develop and validate a decision support model using data mining algorithms, based on morphologic features derived from MRI images, to discriminate between complete responders (CR) and non-complete responders (NCR) patients after neoadjuvant chemoradiotherapy (CRT), in a population of patients with locally advanced rectal cancer (LARC). METHODS: Two populations were retrospectively enrolled: group A (65 patients) was used to train a data mining decision tree algorithm whereas group B (30 patients) was used to validate it. All patients underwent surgery; according to the histology evaluation, patients were divided in CR and NCR. Staging and restaging MRI examinations were retrospectively analysed and seven parameters were considered for data mining classification. Five different classification methods were tested and evaluated in terms of sensitivity, specificity, accuracy and AUC in order to identify the classification model able to achieve the best performance. The best classification algorithm was subsequently applied to group B for validation: sensitivity, specificity, positive and negative predictive value, accuracy and ROC curve were calculated. Inter and intra-reader agreement were calculated. RESULTS: Four features were selected for the development of the classification algorithm: MRI tumor regression grade (MR-TRG), staging volume (SV), tumor volume reduction rate (TVRR) and signal intensity reduction rate (SIRR). The decision tree J48 showed the highest efficiency: when applied to group B, all the CR and 18/21 NCR were correctly classified (sensitivity 85.71%, specificity 100%, PPV 100%, NPV 94.2%, accuracy 95.7%, AUC 0.833). Both inter- and intra-reader evaluation showed good agreement (κ > 0.6). CONCLUSIONS: The proposed decision support model may help in distinguishing between CR and NCR patients with LARC after CRT.

Radiomics in Oncology, Part 1: Technical Principles and Gastrointestinal Application in CT and MRI.

Radiomics has been playing a pivotal role in oncological translational imaging, particularly in cancer diagnosis, prediction prognosis, and therapy response assessment. Recently, promising results were achieved in management of cancer patients by extracting mineable high-dimensional data from medical images, supporting clinicians in decision-making process in the new era of target therapy and personalized medicine. Radiomics could provide quantitative data, extracted from medical images, that could reflect microenvironmental tumor heterogeneity, which might be a useful information for treatment tailoring. Thus, it could be helpful to overcome the main limitations of traditional tumor biopsy, often affected by bias in tumor sampling, lack of repeatability and possible procedure complications. This quantitative approach has been widely investigated as a non-invasive and an objective imaging biomarker in cancer patients; however, it is not applied as a clinical routine due to several limitations related to lack of standardization and validation of images acquisition protocols, features segmentation, extraction, processing, and data analysis. This field is in continuous evolution in each type of cancer, and results support the idea that in the future Radiomics might be a reliable application in oncologic imaging. The first part of this review aimed to describe some radiomic technical principles and clinical applications to gastrointestinal oncologic imaging (CT and MRI) with a focus on diagnosis, prediction prognosis, and assessment of response to therapy.

Radiomics in Oncology, Part 2: Thoracic, Genito-Urinary, Breast, Neurological, Hematologic and Musculoskeletal Applications.

Radiomics has the potential to play a pivotal role in oncological translational imaging, particularly in cancer detection, prognosis prediction and response to therapy evaluation. To date, several studies established Radiomics as a useful tool in oncologic imaging, able to support clinicians in practicing evidence-based medicine, uniquely tailored to each patient and tumor. Mineable data, extracted from medical images could be combined with clinical and survival parameters to develop models useful for the clinicians in cancer patients' assessment. As such, adding Radiomics to traditional subjective imaging may provide a quantitative and extensive cancer evaluation reflecting histologic architecture. In this Part II, we present an overview of radiomic applications in thoracic, genito-urinary, breast, neurological, hematologic and musculoskeletal oncologic applications.

CT for lymph node staging of Colon cancer: not only size but also location and number of lymph node count.

PURPOSE: To evaluate the diagnostic accuracy of imaging features to predict lymph node status of colon cancer using CT. METHODS: This was a retrospective study from 2 tertiary hospitals in South Korea and Netherlands. 317 Colon cancer patients who underwent primary surgical treatment were included. Number of lymph nodes according to the anatomical location, size, cluster, degree of attenuation, shape, presence of internal heterogeneity and ill-defined margin of the lymph node were assessed and compared according to histological lymph node status. RESULTS: The largest short diameter of lymph node and presence of internal heterogeneity of lymph node showed significant association with malignant lymph node status (P < 0.001 and P = 0.041, respectively). The ROC curve analysis revealed AUC of 0.703 for the largest short diameter of lymph node (P < 0.001), and AUC of the presence of internal heterogeneity was 0.630 (P < 0.001). In addition, our study showed that a total number of lymph nodes, regardless of size, (P = 0.022) and number of lymph nodes in peritumoral area (P < 0.001) and along the mesenteric vessels (P < 0.001) on CT demonstrated significant association with malignant status of lymph nodes in colon cancer. CONCLUSIONS: There were significant associations between lymph node status and imaging features of lymph nodes on CT in colon cancer patients. The largest short diameter of lymph node and presence of internal heterogeneity can be used to predict the malignant status of lymph node in colon cancer patients. Also, the number of lymph nodes near the colonic tumor should be considered in assessment of colon cancer lymph node involvement on CT.

3D pelvimetry and biometric measurements: a surgical perspective for colorectal resections.

PURPOSE: Male sex, high BMI, narrow pelvis, and bulky mesorectum were acknowledged as clinical variables correlated with a difficult pelvic dissection in colorectal surgery. This paper aimed at comparing pelvic biometric measurements in female and male patients and at providing a perspective on how pelvimetry segmentation may help in visualizing mesorectal distribution. METHODS: A 3D software was used for segmentation of DICOM data of consecutive patients aged 60 years, who underwent elective abdominal CT scan. The following measurements were estimated: pelvic inlet, outlet, and depth; pubic tubercle height; distances from the promontory to the coccyx and to S3/S4; distance from S3/S4 to coccyx's tip; ischial spines distance; pelvic tilt; offset angle; pelvic inlet angle; angle between the inlet/sacral promontory/coccyx; angle between the promontory/coccyx/pelvic outlet; S3 angle; and pelvic inlet to pelvic depth ratio. The measurements were compared in males and females using statistical analyses. RESULTS: Two-hundred patients (M/F 1:1) were analyzed. Out of 21 pelvimetry measurements, 19 of them documented a significant mean difference between groups. Specifically, female patients had a significantly wider pelvic inlet and outlet but a shorter pelvic depth, and promontory/sacral/coccyx distances, resulting in an augmented inlet/depth ratio when comparing with males (p < 0.0001). The sole exceptions were the straight conjugate (p = 0.06) and S3 angle (p = 0.17). 3D segmentation provided a perspective of the mesorectum distribution according to the pelvic shape. CONCLUSION: Significant differences in the structure of pelvis exist in males and females. Surgeons must be aware of the pelvic shape when approaching the rectum.

Locoregional CT staging of colon cancer: does a learning curve exist?

PURPOSE: To evaluate the learning curve for locoreginal staging of colon cancer in radiologist trainees. METHODS: Eighty-eight cases of colon cancer CT were included in this retrospective study. Four senior radiology residents staged the CTs according to TNM classification. Two out of four radiologists received feedback after reading every 20 cases. Radiologic staging was compared with pathologic staging and the learning curve, diagnostic performance, reader confidence and reading time were evaluated and compared between the two groups (feedback vs. no feedback). Generalized estimating equations logistic regression, QICu statistic, ANOVA and t test/Mann-Whitney test were utilized. RESULTS: Radiologists demonstrated a significant increase in their performance to distinguish between ≤ T2 and ≥ T3 and reached an inflection point at 38 cases, with a significant association with increased number of cases reviewed (P < 0.001). Sensitivity (P < 0.001), specificity (P = 0.030) and NPV (P = 0.002) demonstrated significant associations with increased experience. The overall reader's confidence was significantly higher in the group which received feedback (P < 0.001). There was no significant improvement in performance nor in reader's confidence for N staging (N0 vs. ≥ N1) for all readers. Reading time decreased with experience and showed a significant negative association with experience (P < 0.001). CONCLUSION: Diagnostic performance of senior radiology trainees in differentiating between T2 and T3 colon cancer on CTs improved with increased experience. In contrast, evaluation of lymph node involvement did not improve with more experience. Feedback had no significant effect on improvement of diagnostic performances.

CT texture analysis of liver metastases in PNETs versus NPNETs: Correlation with histopathological findings.

PURPOSE: To compare CT and Texture features of liver metastases in Pancreatic Neuroendocrine Tumors (PNETs) and in Non-Pancreatic Neuroendocrine Tumors (NPNETs) according to tumor grading, overall survival (OS), time to progression (TTP) and Ki67 index. METHODS: 23 patients with PNETs and 25 patients with NPNETs affected by liver metastases were compared. The lesions were G1 and G2 according to WHO classification of tumors. Texture parameters (Mean, Standard Deviation, Entropy, Kurtosis, Skewness, Mean of Positive Pixel) at different spatial scale image filtration (SSF) were evaluated in both arterial and portal phase using a dedicated software for volumetric analysis. All CT images were acquired before the beginning of any medical treatment. RESULTS: The following significant results (P < 0.05) were found: in the arterial phase for value of Skewness between PNETs G2 and NPNETs G2; in the portal phase between PNETs versus NPNETs, PNETs G1 versus NPNETs G1, PNETs G2 versus NPNETs G2; value of Mean in portal phase in PNETs vs NPNETs. Regarding PNETs, a P < 0.05 was found in: inverse correlation between Entropy and TTP; direct correlation between Mean and OS; correlating Kurtosis and high risk of death; correlating Skewness and low risk of death. Regarding NPNETs, P < 0.05 was found in: inverse correlation between Entropy and OS; correlating Entropy and high risk of dying. CONCLUSIONS: This study shows that CT texture features are significantly different in PNETs from NPNETs. Additionally, textural features such as Entropy, Kurtosis and Skewness, were found to have significant correlation with higher mortality risk.