Machine Learning Detection and Characterization of Splenic Injuries on Abdominal Computed Tomography.

BACKGROUND: Multi-detector contrast-enhanced abdominal computed tomography (CT) allows for the accurate detection and classification of traumatic splenic injuries, leading to improved patient management. Their effective use requires rapid study interpretation, which can be a challenge on busy emergency radiology services. A machine learning system has the potential to automate the process, potentially leading to a faster clinical response. This study aimed to create such a system. METHOD: Using the American Association for the Surgery of Trauma (AAST), spleen injuries were classified into 3 classes: normal, low-grade (AAST grade I-III) injuries, and high-grade (AAST grade IV and V) injuries. Employing a 2-stage machine learning strategy, spleens were initially segmented from input CT images and subsequently underwent classification via a 3D dense convolutional neural network (DenseNet). RESULTS: This single-centre retrospective study involved trauma protocol CT scans performed between January 1, 2005, and July 31, 2021, totaling 608 scans with splenic injuries and 608 without. Five board-certified fellowship-trained abdominal radiologists utilizing the AAST injury scoring scale established ground truth labels. The model achieved AUC values of 0.84, 0.69, and 0.90 for normal, low-grade injuries, and high-grade splenic injuries, respectively. CONCLUSIONS: Our findings demonstrate the feasibility of automating spleen injury detection using our method with potential applications in improving patient care through radiologist worklist prioritization and injury stratification. Future endeavours should concentrate on further enhancing and optimizing our approach and testing its use in a real-world clinical environment.

Radiomic Analysis: Study Design, Statistical Analysis, and Other Bias Mitigation Strategies.

Rapid advances in automated methods for extracting large numbers of quantitative features from medical images have led to tremendous growth of publications reporting on radiomic analyses. Translation of these research studies into clinical practice can be hindered by biases introduced during the design, analysis, or reporting of the studies. Herein, the authors review biases, sources of variability, and pitfalls that frequently arise in radiomic research, with an emphasis on study design and statistical analysis considerations. Drawing on existing work in the statistical, radiologic, and machine learning literature, approaches for avoiding these pitfalls are described.

Genetic Determinants of Outcome in Intrahepatic Cholangiocarcinoma.

BACKGROUND AND AIM: Genetic alterations in intrahepatic cholangiocarcinoma (iCCA) are increasingly well characterized, but their impact on outcome and prognosis remains unknown. APPROACH AND RESULTS: This bi-institutional study of patients with confirmed iCCA (n = 412) used targeted next-generation sequencing of primary tumors to define associations among genetic alterations, clinicopathological variables, and outcome. The most common oncogenic alterations were isocitrate dehydrogenase 1 (IDH1; 20%), AT-rich interactive domain-containing protein 1A (20%), tumor protein P53 (TP53; 17%), cyclin-dependent kinase inhibitor 2A (CDKN2A; 15%), breast cancer 1-associated protein 1 (15%), FGFR2 (15%), polybromo 1 (12%), and KRAS (10%). IDH1/2 mutations (mut) were mutually exclusive with FGFR2 fusions, but neither was associated with outcome. For all patients, TP53 (P < 0.0001), KRAS (P = 0.0001), and CDKN2A (P < 0.0001) alterations predicted worse overall survival (OS). These high-risk alterations were enriched in advanced disease but adversely impacted survival across all stages, even when controlling for known correlates of outcome (multifocal disease, lymph node involvement, bile duct type, periductal infiltration). In resected patients (n = 209), TP53mut (HR, 1.82; 95% CI, 1.08-3.06; P = 0.03) and CDKN2A deletions (del; HR, 3.40; 95% CI, 1.95-5.94; P < 0.001) independently predicted shorter OS, as did high-risk clinical variables (multifocal liver disease [P < 0.001]; regional lymph node metastases [P < 0.001]), whereas KRASmut (HR, 1.69; 95% CI, 0.97-2.93; P = 0.06) trended toward statistical significance. The presence of both or neither high-risk clinical or genetic factors represented outcome extremes (median OS, 18.3 vs. 74.2 months; P < 0.001), with high-risk genetic alterations alone (median OS, 38.6 months; 95% CI, 28.8-73.5) or high-risk clinical variables alone (median OS, 37.0 months; 95% CI, 27.6-not available) associated with intermediate outcome. TP53mut, KRASmut, and CDKN2Adel similarly predicted worse outcome in patients with unresectable iCCA. CDKN2Adel tumors with high-risk clinical features were notable for limited survival and no benefit of resection over chemotherapy. CONCLUSIONS: TP53, KRAS, and CDKN2A alterations were independent prognostic factors in iCCA when controlling for clinical and pathologic variables, disease stage, and treatment. Because genetic profiling can be integrated into pretreatment therapeutic decision-making, combining clinical variables with targeted tumor sequencing may identify patient subgroups with poor outcome irrespective of treatment strategy.

Recurrence After Resection of Pancreatic Cancer: Can Radiomics Predict Patients at Greatest Risk of Liver Metastasis?

BACKGROUND: Liver metastasis (LM) after pancreatic ductal adenocarcinoma (PDAC) resection is common but difficult to predict and has grave prognosis. We combined preoperative clinicopathological variables and quantitative analysis of computed tomography (CT) imaging to predict early LM. METHODS: We retrospectively evaluated patients with PDAC submitted to resection between 2005 and 2014 and identified clinicopathological variables associated with early LM. We performed liver radiomic analysis on preoperative contrast-enhanced CT scans and developed a logistic regression classifier to predict early LM (< 6 months). RESULTS: In 688 resected PDAC patients, there were 516 recurrences (75%). The cumulative incidence of LM at 5 years was 41%, and patients who developed LM first (n = 194) had the lowest 1-year overall survival (OS) (34%), compared with 322 patients who developed extrahepatic recurrence first (61%). Independent predictors of time to LM included poor tumor differentiation (hazard ratio (HR) = 2.30; P < 0.001), large tumor size (HR = 1.17 per 2-cm increase; P = 0.048), lymphovascular invasion (HR = 1.50; P = 0.015), and liver Fibrosis-4 score (HR = 0.89 per 1-unit increase; P = 0.029) on multivariate analysis. A model using radiomic variables that reflect hepatic parenchymal heterogeneity identified patients at risk for early LM with an area under the receiver operating characteristic curve (AUC) of 0.71; the performance of the model was improved by incorporating preoperative clinicopathological variables (tumor size and differentiation status; AUC = 0.74, negative predictive value (NPV) = 0.86). CONCLUSIONS: We confirm the adverse survival impact of early LM after resection of PDAC. We further show that a model using radiomic data from preoperative imaging combined with tumor-related variables has great potential for identifying patients at high risk for LM and may help guide treatment selection.

Patterns of Metastatic Disease in Patients with Cancer Derived from Natural Language Processing of Structured CT Radiology Reports over a 10-year Period.

Background Patterns of metastasis in cancer are increasingly relevant to prognostication and treatment planning but have historically been documented by means of autopsy series. Purpose To show the feasibility of using natural language processing (NLP) to gather accurate data from radiology reports for assessing spatial and temporal patterns of metastatic spread in a large patient cohort. Materials and Methods In this retrospective longitudinal study, consecutive patients who underwent CT from July 2009 to April 2019 and whose CT reports followed a departmental structured template were included. Three radiologists manually curated a sample of 2219 reports for the presence or absence of metastases across 13 organs; these manually curated reports were used to develop three NLP models with an 80%-20% split for training and test sets. A separate random sample of 448 manually curated reports was used for validation. Model performance was measured by accuracy, precision, and recall for each organ. The best-performing NLP model was used to generate a final database of metastatic disease across all patients. For each cancer type, statistical descriptive reports were provided by analyzing the frequencies of metastatic disease at the report and patient levels. Results In 91 665 patients (mean age ± standard deviation, 61 years ± 15; 46 939 women), 387 359 reports were labeled. The best-performing NLP model achieved accuracies from 90% to 99% across all organs. Metastases were most frequently reported in abdominopelvic (23.6% of all reports) and thoracic (17.6%) nodes, followed by lungs (14.7%), liver (13.7%), and bones (9.9%). Metastatic disease tropism is distinct among common cancers, with the most common first site being bones in prostate and breast cancers and liver among pancreatic and colorectal cancers. Conclusion Natural language processing may be applied to cancer patients' CT reports to generate a large database of metastatic phenotypes. Such a database could be combined with genomic studies and used to explore prognostic imaging phenotypes with relevance to treatment planning. © RSNA, 2021 Online supplemental material is available for this article.

Differences in Liver Parenchyma are Measurable with CT Radiomics at Initial Colon Resection in Patients that Develop Hepatic Metastases from Stage II/III Colon Cancer.

BACKGROUND: Currently, there are no methods to identify patients with an increased risk of liver metastases to guide patient selection for liver-directed therapies. We tried to determine whether quantitative image features (radiomics) of the liver obtained from preoperative staging CT scans at the time of initial colon resection differ in patients that subsequently develop liver metastases, extrahepatic metastases, or demonstrate prolonged disease-free survival. METHODS: Patients who underwent resection of stage II/III colon cancer from 2004 to 2012 with available preoperative CT scans were included in this single-institution, retrospective case-control study. Patients were grouped by initial recurrence patterns: liver recurrence, extrahepatic recurrence, or no evidence of disease at 5 years. Radiomic features of the liver parenchyma extracted from CT images were compared across groups. RESULTS: The cohort consisted of 120 patients divided evenly between three recurrence groups, with an equal number of stage II and III patients in each group. After adjusting for multiple comparisons, 44 of 254 (17%) imaging features displayed different distributions across the three patient groups (p < 0.05), with the clearest distinction between those with liver recurrence and no evidence of disease. Increased heterogeneity in the liver parenchyma by radiomic analysis was protective of liver metastases. CONCLUSIONS: CT radiomics is a promising tool to identify patients at high risk of developing liver metastases and is worthy of further investigation and validation.

Radiomic feature reproducibility in contrast-enhanced CT of the pancreas is affected by variabilities in scan parameters and manual segmentation.

OBJECTIVES: This study aims to measure the reproducibility of radiomic features in pancreatic parenchyma and ductal adenocarcinomas (PDAC) in patients who underwent consecutive contrast-enhanced computed tomography (CECT) scans. METHODS: In this IRB-approved and HIPAA-compliant retrospective study, 37 pairs of scans from 37 unique patients who underwent CECTs within a 2-week interval were included in the analysis of the reproducibility of features derived from pancreatic parenchyma, and a subset of 18 pairs of scans were further analyzed for the reproducibility of features derived from PDAC. In each patient, pancreatic parenchyma and pancreatic tumor (when present) were manually segmented by two radiologists independently. A total of 266 radiomic features were extracted from the pancreatic parenchyma and tumor region and also the volume and diameter of the tumor. The concordance correlation coefficient (CCC) was calculated to assess feature reproducibility for each patient in three scenarios: (1) different radiologists, same CECT; (2) same radiologist, different CECTs; and (3) different radiologists, different CECTs. RESULTS: Among pancreatic parenchyma-derived features, using a threshold of CCC > 0.90, 58/266 (21.8%) and 48/266 (18.1%) features met the threshold for scenario 1, 14/266 (5.3%) and 15/266 (5.6%) for scenario 2, and 14/266 (5.3%) and 10/266 (3.8%) for scenario 3. Among pancreatic tumor-derived features, 11/268 (4.1%) and 17/268 (6.3%) features met the threshold for scenario 1, 1/268 (0.4%) and 5/268 (1.9%) features met the threshold for scenario 2, and no features for scenario 3 met the threshold, respectively. CONCLUSIONS: Variations between CECT scans affected radiomic feature reproducibility to a greater extent than variation in segmentation. A smaller number of pancreatic tumor-derived radiomic features were reproducible compared with pancreatic parenchyma-derived radiomic features under the same conditions. KEY POINTS: • For pancreatic-derived radiomic features from contrast-enhanced CT (CECT), fewer than 25% are reproducible (with a threshold of CCC < 0.9) in a clinical heterogeneous dataset. • Variations between CECT scans affected the number of reproducible radiomic features to a greater extent than variations in radiologist segmentation. • A smaller number of pancreatic tumor-derived radiomic features were reproducible compared with pancreatic parenchyma-derived radiomic features under the same conditions.

Regional differences in gallbladder cancer pathogenesis: Insights from a multi-institutional comparison of tumor mutations.

BACKGROUND: Although rare in the United States, gallbladder cancer (GBCA) is a common cause of cancer death in some parts of the world. To investigate regional differences in pathogenesis and outcomes for GBCA, tumor mutations were analyzed from a sampling of specimens. METHODS: Primary tumors from patients with GBCA who were treated in Chile, Japan, and the United States between 1999 and 2016 underwent targeted sequencing of known cancer-associated genes. Fisher exact and Kruskal-Wallis tests assessed differences in clinicopathologic and genetic factors. Kaplan-Meier methods evaluated differences in overall survival from the time of surgery between mutations. RESULTS: A total of 81 patients were included. Japanese patients (11 patients) were older (median age, 72 years [range, 54-81 years]) compared with patients from Chile (21 patients; median age, 59 years [range, 32-73 years]) and the United States (49 patients; median age, 66 years [range, 46-87 years]) (P = .002) and had more well-differentiated tumors (46% vs 0% for Chile/United States; P < .001) and fewer gallstone-associated cancers (36% vs 67% for Chile and 69% for the United States; P = .13). Japanese patients had a median mutation burden of 6 (range, 1-23) compared with Chile (median mutation burden, 7 [range, 3-20]) and the United States (median mutation burden, 4 [range, 0-27]) (P = .006). Tumors from Japanese patients lacked AT-rich interaction domain 1A (ARID1A) and phosphatidylinositol-4,5-bisphosphate 3-kinase catalytic subunit alpha (PIK3CA) mutations, whereas Chilean tumors lacked Erb-B2 receptor tyrosine kinase 3 (ERBB3) and AT-rich interaction domain 2 (ARID2) mutations. SMAD family member 4 (SMAD4) was found to be mutated similarly across centers (38% in Chile, 36% in Japan, and 27% in the United States; P = .68) and was univariately associated with worse overall survival (median, 10 months vs 25 months; P = .039). At least one potentially actionable gene was found to be altered in 80% of tumors. CONCLUSIONS: Differences in clinicopathologic variables suggest the possibility of distinct GBCA pathogenesis in Japanese patients, which may be supported by differences in mutation pattern. Among all centers, SMAD4 mutations were detected in approximately one-third of patients and may represent a converging factor associated with worse survival. The majority of patients carried mutations in actionable gene targets, which may inform the design of future trials.

Quantitative imaging features of pretreatment CT predict volumetric response to chemotherapy in patients with colorectal liver metastases.

OBJECTIVES: This study investigates whether quantitative image analysis of pretreatment CT scans can predict volumetric response to chemotherapy for patients with colorectal liver metastases (CRLM). METHODS: Patients treated with chemotherapy for CRLM (hepatic artery infusion (HAI) combined with systemic or systemic alone) were included in the study. Patients were imaged at baseline and approximately 8 weeks after treatment. Response was measured as the percentage change in tumour volume from baseline. Quantitative imaging features were derived from the index hepatic tumour on pretreatment CT, and features statistically significant on univariate analysis were included in a linear regression model to predict volumetric response. The regression model was constructed from 70% of data, while 30% were reserved for testing. Test data were input into the trained model. Model performance was evaluated with mean absolute prediction error (MAPE) and R2. Clinicopatholologic factors were assessed for correlation with response. RESULTS: 157 patients were included, split into training (n = 110) and validation (n = 47) sets. MAPE from the multivariate linear regression model was 16.5% (R2 = 0.774) and 21.5% in the training and validation sets, respectively. Stratified by HAI utilisation, MAPE in the validation set was 19.6% for HAI and 25.1% for systemic chemotherapy alone. Clinical factors associated with differences in median tumour response were treatment strategy, systemic chemotherapy regimen, age and KRAS mutation status (p < 0.05). CONCLUSION: Quantitative imaging features extracted from pretreatment CT are promising predictors of volumetric response to chemotherapy in patients with CRLM. Pretreatment predictors of response have the potential to better select patients for specific therapies. KEY POINTS: • Colorectal liver metastases (CRLM) are downsized with chemotherapy but predicting the patients that will respond to chemotherapy is currently not possible. • Heterogeneity and enhancement patterns of CRLM can be measured with quantitative imaging. • Prediction model constructed that predicts volumetric response with 20% error suggesting that quantitative imaging holds promise to better select patients for specific treatments.

Preoperative risk prediction for intraductal papillary mucinous neoplasms by quantitative CT image analysis.

BACKGROUND: Intraductal papillary mucinous neoplasms (IPMNs) are radiographically identifiable potential precursor lesions of pancreatic adenocarcinoma. While resection is recommended when main duct dilation is present, management of branch duct IPMN (BD-IPMN) remains controversial. This study sought to evaluate whether preoperative quantitative imaging features of BD-IPMNs could distinguish low-risk disease (low- and intermediate-grade dysplasia) from high-risk disease (high-grade dysplasia and invasive carcinoma). METHODS: Patients who underwent resection between 2005 and 2015 with pathologically proven BD-IPMN and a preoperative CT scan were included in the study. Quantitative image features were extracted using texture analysis and a novel quantitative mural nodularity feature developed for the study. Significant features on univariate analysis were combined with clinical variables to build a multivariate prediction model. RESULTS: Within the study group of 103 patients, 76 (74%) had low-risk disease and 27 (26%) had high-risk disease. Quantitative imaging features were prognostic of low-vs. high-risk disease. The model based only on clinical variables achieved an AUC of 0.67 and 0.79 with the addition of quantitative imaging features. CONCLUSION: Quantitative image analysis of BD-IPMNs is a novel method that may enable risk stratification. External validation may provide a reliable non-invasive prognostic tool for clinicians.

Survival Prediction in Pancreatic Ductal Adenocarcinoma by Quantitative Computed Tomography Image Analysis.

BACKGROUND: Pancreatic cancer is a highly lethal cancer with no established a priori markers of survival. Existing nomograms rely mainly on post-resection data and are of limited utility in directing surgical management. This study investigated the use of quantitative computed tomography (CT) features to preoperatively assess survival for pancreatic ductal adenocarcinoma (PDAC) patients. METHODS: A prospectively maintained database identified consecutive chemotherapy-naive patients with CT angiography and resected PDAC between 2009 and 2012. Variation in CT enhancement patterns was extracted from the tumor region using texture analysis, a quantitative image analysis tool previously described in the literature. Two continuous survival models were constructed, with 70% of the data (training set) using Cox regression, first based only on preoperative serum cancer antigen (CA) 19-9 levels and image features (model A), and then on CA19-9, image features, and the Brennan score (composite pathology score; model B). The remaining 30% of the data (test set) were reserved for independent validation. RESULTS: A total of 161 patients were included in the analysis. Training and test sets contained 113 and 48 patients, respectively. Quantitative image features combined with CA19-9 achieved a c-index of 0.69 [integrated Brier score (IBS) 0.224] on the test data, while combining CA19-9, imaging, and the Brennan score achieved a c-index of 0.74 (IBS 0.200) on the test data. CONCLUSION: We present two continuous survival prediction models for resected PDAC patients. Quantitative analysis of CT texture features is associated with overall survival. Further work includes applying the model to an external dataset to increase the sample size for training and to determine its applicability.

3D image guidance assisted identification of colorectal cancer liver metastases not seen on intraoperative ultrasound: results from a prospective trial.

BACKGROUND: Neoadjuvant treatment of colorectal liver metastases has become increasingly common, and while effective, often renders small metastases difficult to visualize on intraoperative US. The objective of this study was to determine the utility of a 3D image-guidance system in patients with intraoperative sonographically-occult CRLM. METHODS: 50 patients with at least one CRLM ≤ 1.5 cm were enrolled in this prospective trial of an FDA-approved Explorer image-guidance system. If the tumor(s) seen on preoperative imaging were not identified with intraoperative US, Explorer was used to target the US examination to the involved area for a more focused assessment. The primary endpoint was the proportion of cases with sonographically-occult metastases identified using Explorer. RESULTS: Forty-eight patients with preoperative scans within eight weeks of surgery were included for analysis. Forty-six patients were treated with preoperative chemotherapy (median 4 months, range 2-24 months). Overall, 22 sonographically-occult tumors in 14 patients were interrogated by Explorer, of which 15 tumors in 10 patients were located with image-guidance assistance. The only difference between patients with tumors not identified on US and those who did was the number of tumors (median 3 vs. 2, p = 0.018). CONCLUSION: 3D image-guidance can assist in identifying small CRLM, particularly after treatment with chemotherapy. TRIAL REGISTRATION: ClinicalTrials.gov, NCT02806037, https://clinicaltrials.gov/ct2/show/NCT02806037.

Computed Tomography Image Texture: A Noninvasive Prognostic Marker of Hepatic Recurrence After Hepatectomy for Metastatic Colorectal Cancer.

BACKGROUND: Recurrence after resection of colorectal liver metastases (CRLMs) occurs in up to 75% of patients. Preoperative prediction of hepatic recurrence may inform therapeutic strategies at the time of initial resection. Texture analysis (TA) is an established technique that quantifies pixel intensity variations (heterogeneity) on cross-sectional imaging. We hypothesized that tumoral and parenchymal changes that are predictive of overall survival (OS) and recurrence in the future liver remnant (FLR) can be detected using TA on preoperative computed tomography (CT) images. METHODS: Patients who underwent resection for CRLM between 2003 and 2007 with appropriate preoperative CT scans were included (n = 198) in this retrospective study. Texture features extracted from the tumor and FLR, and clinicopathologic variables, were incorporated into a multivariable survival model. RESULTS: Quantitative imaging features of the FLR were an independent predictor of both OS and hepatic disease-free survival (HDFS). Tumor texture showed significant association with OS. TA of the FLR allowed patient stratification into two groups, with significantly different risks of hepatic recurrence (hazard ratio 2.09, 95% confidence interval 1.33-3.28; p = 0.001). Patients with homogeneous parenchyma had approximately twice the risk of hepatic recurrence (41 vs. 20%). CONCLUSION: TA of the tumor and FLR are independently associated with OS, and TA of the FLR is independently associated with HDFS. Patients with homogeneous parenchyma had a significantly higher risk of hepatic recurrence. Preoperative TA of the liver represents a potential biomarker to identify patients at risk of liver recurrence after resection for CRLM.

Assessing splenic enlargement on CT by unidimensional measurement changes in patients with colorectal liver metastases.

PURPOSE: The aim of this study was to assess splenic volume and to correlate unidimensional measurements with reference volumetric changes in chemotherapy-treated patients with colorectal cancer (CRC) liver metastases. METHODS: Forty consecutive patients were selected from the cohort of a previously reported study of chemotherapy-related morbidity following major hepatectomy for CRC liver metastases. Patients were treated for 6 months prior to resection, with imaging performed at baseline and after 6 months of chemotherapy. Three unidimensional spleen measurements were recorded-width, thickness, and height (W, T, and H). Reference splenic volume was measured at baseline and after chemotherapy. The best unidimensional splenic measurement was determined by regression analysis. The 95% CI for the predicted values and R (2) values was calculated for each regression. The percentage of volume increase at 6 months was calculated. RESULTS: W and H showed the highest correlation with splenic volume prior to and following chemotherapy (R (2) = 0.65-0.74, p < 0.001), while T showed a low correlation (R (2) = 0.11 and 0.18, p < 0.05). The mean reference splenic volume increased after 6 months of chemotherapy compared to baseline (326 vs. 278 mL). Splenic volume changes showed the highest correlation with changes in W (R (2) = 0.56, p < 0.001), then H (R (2) = 0.40, p < 0.001), but were not significantly correlated with changes in T (R (2) = 0.01, p = 0.055). CONCLUSIONS: Our results show the potential utility of measuring changes in splenic width to predict clinically significant changes in splenic volume in chemotherapy-treated patients with CRC liver metastases.

Early trends in serum phosphate and creatinine levels are associated with mortality following major hepatectomy.

BACKGROUND: Mortality after major hepatectomy remains high and is frequently related to post-hepatectomy liver failure (PHLF). Other than pre-existing liver disease and a small future liver remnant, few patient factors or early postoperative indicators identify patients at elevated risk for PHLF and mortality. METHODS: Data on demographics, comorbidities, operative procedures and postoperative laboratory trends were reviewed for patients submitted to major hepatectomy (at least three Couinaud segments) for malignancy during 1998-2013. These factors were compared among patients who died within 90 days, survivors who met the 50-50 criteria and all remaining survivors. RESULTS: A total of 1528 patients underwent major hepatectomy during the study period. Of these, 947 had metastatic colorectal cancer and underwent resection of a median of four segments. Overall, 49 patients (3.2%) died within 90 days of surgery and 48 patients (3.1%) met the 50-50 criteria for PHLF; 30 of these patients survived 90 days. Operative blood loss was higher in patients who died within 90 days compared with survivors (1.0 l versus 0.5 l; P < 0.001). Despite equivalent perioperative resuscitation and urine output, non-survivors had higher creatinine and phosphate levels than survivors on postoperative day (PoD) 1 (1.1 mg/dl versus 0.9 mg/dl and 4.6 mg/dl versus 3.7 mg/dl, respectively; P < 0.001). CONCLUSIONS: Early trends in creatinine and phosphate (between the day of surgery and PoD 1) identify patients at risk for PHLF and mortality.

Medical image synthesis via conditional GANs: Application to segmenting brain tumours.

Accurate brain tumour segmentation is critical for tasks such as surgical planning, diagnosis, and analysis, with magnetic resonance imaging (MRI) being the preferred modality due to its excellent visualisation of brain tissues. However, the wide intensity range of voxel values in MR scans often results in significant overlap between the density distributions of different tumour tissues, leading to reduced contrast and segmentation accuracy. This paper introduces a novel framework based on conditional generative adversarial networks (cGANs) aimed at enhancing the contrast of tumour subregions for both voxel-wise and region-wise segmentation approaches. We present two models: Enhancement and Segmentation GAN (ESGAN), which combines classifier loss with adversarial loss to predict central labels of input patches, and Enhancement GAN (EnhGAN), which generates high-contrast synthetic images with reduced inter-class overlap. These synthetic images are then fused with corresponding modalities to emphasise meaningful tissues while suppressing weaker ones. We also introduce a novel generator that adaptively calibrates voxel values within input patches, leveraging fully convolutional networks. Both models employ a multi-scale Markovian network as a GAN discriminator to capture local patch statistics and estimate the distribution of MR images in complex contexts. Experimental results on publicly available MR brain tumour datasets demonstrate the competitive accuracy of our models compared to current brain tumour segmentation techniques.

Simulated deep CT characterization of liver metastases with high-resolution filtered back projection reconstruction.

Early diagnosis and accurate prognosis of colorectal cancer is critical for determining optimal treatment plans and maximizing patient outcomes, especially as the disease progresses into liver metastases. Computed tomography (CT) is a frontline tool for this task; however, the preservation of predictive radiomic features is highly dependent on the scanning protocol and reconstruction algorithm. We hypothesized that image reconstruction with a high-frequency kernel could result in a better characterization of liver metastases features via deep neural networks. This kernel produces images that appear noisier but preserve more sinogram information. A simulation pipeline was developed to study the effects of imaging parameters on the ability to characterize the features of liver metastases. This pipeline utilizes a fractal approach to generate a diverse population of shapes representing virtual metastases, and then it superimposes them on a realistic CT liver region to perform a virtual CT scan using CatSim. Datasets of 10,000 liver metastases were generated, scanned, and reconstructed using either standard or high-frequency kernels. These data were used to train and validate deep neural networks to recover crafted metastases characteristics, such as internal heterogeneity, edge sharpness, and edge fractal dimension. In the absence of noise, models scored, on average, 12.2% ( α = 0.012 ) and 7.5% ( α = 0.049 ) lower squared error for characterizing edge sharpness and fractal dimension, respectively, when using high-frequency reconstructions compared to standard. However, the differences in performance were statistically insignificant when a typical level of CT noise was simulated in the clinical scan. Our results suggest that high-frequency reconstruction kernels can better preserve information for downstream artificial intelligence-based radiomic characterization, provided that noise is limited. Future work should investigate the information-preserving kernels in datasets with clinical labels.

Use of Response Permutation to Measure an Imaging Dataset's Susceptibility to Overfitting by Selected Standard Analysis Pipelines.

RATIONALE AND OBJECTIVES: This study demonstrates a method for quantifying the impact of overfitting on the receiving operator characteristic curve (AUC) when using standard analysis pipelines to develop imaging biomarkers. We illustrate the approach using two publicly available repositories of radiology and pathology images for breast cancer diagnosis. MATERIALS AND METHODS: For each dataset, we permuted the outcome (cancer diagnosis) values to eliminate any true association between imaging features and outcome. Seven types of classification models (logistic regression, linear discriminant analysis, Naïve Bayes, linear support vector machines, nonlinear support vector machine, random forest, and multi-layer perceptron) were fitted to each scrambled dataset and evaluated by each of four techniques (all data, hold-out, 10-fold cross-validation, and bootstrapping). After repeating this process for a total of 50 outcome permutations, we averaged the resulting AUCs. Any increase over a null AUC of 0.5 can be attributed to overfitting. RESULTS: Applying this approach and varying sample size and the number of imaging features, we found that failing to control for overfitting could result in near-perfect prediction (AUC near 1.0). Cross-validation offered greater protection against overfitting than the other evaluation techniques, and for most classification algorithms a sample size of at least 200 was required to assess as few as 10 features with less than 0.05 AUC inflation attributable to overfitting. CONCLUSION: This approach could be applied to any curated dataset to suggest the number of features and analysis approaches to limit overfitting.