Artificial Intelligence for Quantifying Cumulative Small Bowel Disease Severity on CT-Enterography in Crohn's Disease.

INTRODUCTION: Assessing the cumulative degree of bowel injury in ileal Crohn's disease (CD) is difficult. We aimed to develop machine learning (ML) methodologies for automated estimation of cumulative ileal injury on computed tomography-enterography (CTE) to help predict future bowel surgery. METHODS: Adults with ileal CD using biologic therapy at a tertiary care center underwent ML analysis of CTE scans. Two fellowship-trained radiologists graded bowel injury severity at granular spatial increments along the ileum (1 cm), called mini-segments. ML segmentation methods were trained on radiologist grading with predicted severity and then spatially mapped to the ileum. Cumulative injury was calculated as the sum (S-CIDSS) and mean of severity grades along the ileum. Multivariate models of future small bowel resection were compared with cumulative ileum injury metrics and traditional bowel measures, adjusting for laboratory values, medications, and prior surgery at the time of CTE. RESULTS: In 229 CTE scans, 8,424 mini-segments underwent analysis. Agreement between ML and radiologists injury grading was strong (κ = 0.80, 95% confidence interval 0.79-0.81) and similar to inter-radiologist agreement (κ = 0.87, 95% confidence interval 0.85-0.88). S-CIDSS (46.6 vs 30.4, P = 0.0007) and mean cumulative injury grade scores (1.80 vs 1.42, P < 0.0001) were greater in CD biologic users that went to future surgery. Models using cumulative spatial metrics (area under the curve = 0.76) outperformed models using conventional bowel measures, laboratory values, and medical history (area under the curve = 0.62) for predicting future surgery in biologic users. DISCUSSION: Automated cumulative ileal injury scores show promise for improving prediction of outcomes in small bowel CD. Beyond replicating expert judgment, spatial enterography analysis can augment the personalization of bowel assessment in CD.

Incorporation of quantitative imaging data using artificial intelligence improves risk prediction in veterans with liver disease.

BACKGROUND AND AIMS: Utilization of electronic health records data to derive predictive indexes such as the electronic Child-Turcotte-Pugh (eCTP) Score can have significant utility in health care delivery. Within the records, CT scans contain phenotypic data which have significant prognostic value. However, data extractions have not traditionally been applied to imaging data. In this study, we used artificial intelligence to automate biomarker extraction from CT scans and examined the value of these features in improving risk prediction in patients with liver disease. APPROACH AND RESULTS: Using a regional liver disease cohort from the Veterans Health System, we retrieved administrative, laboratory, and clinical data for Veterans who had CT scans performed for any clinical indication between 2008 and 2014. Imaging biomarkers were automatically derived using the analytic morphomics platform. In all, 4614 patients were included. We found that the eCTP Score had a Concordance index of 0.64 for the prediction of overall mortality while the imaging-based model alone or with eCTP Score performed significantly better [Concordance index of 0.72 and 0.73 ( p <0.001)]. For the subset of patients without hepatic decompensation at baseline (n=4452), the Concordance index for predicting future decompensation was 0.67, 0.79, and 0.80 for eCTP Score, imaging alone, or combined, respectively. CONCLUSIONS: This proof of concept demonstrates that the potential of utilizing automated extraction of imaging features within CT scans either alone or in conjunction with classic health data can improve risk prediction in patients with chronic liver disease.

Early Development of the Mouse Morphome.

INTRODUCTION: Analytical morphomics focuses on extracting objective and quantifiable data from clinical computed tomography (CT) scans to measure patients' frailty. Studies are currently retrospective in nature; therefore, it would be beneficial to develop animal models for well-controlled, prospective studies. The aim of this study is to develop an in vivo microCT protocol for the longitudinal acquisition of whole-body images suitable for morphomic analyses of bone. METHODS: The authors performed phantom studies on 2 microCT systems (Inveon and CT120) to study tissue radiodensity and further characterize system performance for collecting animal data. The authors also describe their design of a phantom-immobilization device using phantoms and an ovariectomized (OVX) mouse. RESULTS: The authors discovered increased consistency along the z-axis for scans acquired on the Inveon compared with CT120, and calibration by individual slice reduces variability. Objects in the field of view had more impact on measurement acquired using the CT120 compared with the Inveon. The authors also found that using the middle 80% of slices for data analysis further decreased variability, on both systems. Moreover, bone-mineral-density calibration using the QCT Pro Mini phantom improved bone-mineral-density estimates across energy spectra, which helped confirm our technique. Comparison of weekly body weights and terminal uterine mass between sham and OVX groups validated our model. DISCUSSION: The authors present a refined microCT protocol to collect reliable and objective data. This data will be used to establish a platform for research animal morphomics that can be used to test hypotheses developed from clinical human morphomics.

Use of analytic morphomics of liver, spleen, and body composition to identify patients at risk for cirrhosis.

BACKGROUND & AIMS: A diagnosis of cirrhosis can be made on the basis of findings from imaging studies, but these are subjective. Analytic morphomics uses computational image processing algorithms to provide precise and detailed measurements of organs and body tissues. We investigated whether morphomic parameters can be used to identify patients with cirrhosis. METHODS: In a retrospective study, we performed analytic morphomics on data collected from 357 patients evaluated at the University of Michigan from 2004 to 2012 who had a liver biopsy within 6 months of a computed tomography scan for any reason. We used logistic regression with elastic net regularization and cross-validation to develop predictive models for cirrhosis, within 80% randomly selected internal training set. The other 20% data were used as internal test set to ensure that model overfitting did not occur. In validation studies, we tested the performance of our models on an external cohort of patients from a different health system. RESULTS: Our predictive models, which were based on analytic morphomics and demographics (morphomics model) or analytic morphomics, demographics, and laboratory studies (full model), identified patients with cirrhosis with area under the receiver operating characteristic curve (AUROC) values of 0.91 and 0.90, respectively, compared with 0.69, 0.77, and 0.76 for aspartate aminotransferase-to-platelet ratio, Lok Score, and FIB-4, respectively, by using the same data set. In the validation set, our morphomics model identified patients who developed cirrhosis with AUROC value of 0.97, and the full model identified them with AUROC value of 0.90. CONCLUSIONS: We used analytic morphomics to demonstrate that cirrhosis can be objectively quantified by using medical imaging. In a retrospective analysis of multi-protocol scans, we found that it is possible to identify patients who have cirrhosis on the basis of analyses of preexisting scans, without significant additional risk or cost.

Abdominal wall dynamics after component separation hernia repair.

BACKGROUND: The component separation technique (CST) is an important technique now used frequently in complex ventral hernia repair (VHR). Although this technique has demonstrated superior success rates, there is a paucity of research describing how release of the external obliques coupled with rectus myofascial advancement alters the morphology of the abdominal architecture. In this study, we apply the new concept of analytic morphomics to describe the immediate changes in morphology of the abdomen that take place after VHR by CST. METHODS: We identified 21 patients who underwent VHR by CST and received both preoperative and postoperative computed tomography scans between 2004 and 2009 in our clinical database. The surgical technique involved incisional release of the external oblique muscle lateral to the linea semilunaris with rectus abdominis myofascial advancement in all patients. Using semiautomated morphomic analysis, we measured the pre- and post-operative dimensions of the abdominal wall including the anterior-posterior distance from the anterior vertebra-to-skin and fascia along with the circumferential area of the skin and fascial compartments. Paired Student t-tests were used to compare pre- and post-operative values. RESULTS: After hernia repair, there was a decrease in the anterior vertebra-to-skin distance (16.6 cm-15.8 cm, P = 0.007). There were also decreases in total body area (968.0 cm(2)-928.6 cm(2), P = 0.017) and total body circumference (113.6 cm-111.4 cm, P = 0.016). The distance from fascia to skin decreased as well, almost to the point of statistical significance (3.3 cm-2.9 cm, P = 0.0505). Interestingly, fascia area and circumference did not decrease significantly after the operation (578.2 cm(2)-572.5 cm(2), P = 0.519, and 89.1 cm-88.6 cm, P = 0.394, respectively). CONCLUSIONS: Morphomic analysis can be used to compare and pre- and post-operative changes in patients undergoing abdominal surgery. Our study demonstrates that component separation affects the dimensions of the entire abdomen, but leaves the fascia area and circumference relatively unchanged. These changes in the abdominal wall may help explain the muscular changes observed as a result of this operation and demonstrate that this is a functional operation that restores fascial area. By better defining the effects of this procedure, we can better understand the reason for its clinical success.

Use of temporal morphomic indices as a clinically important variable in the diagnosis of nonsyndromic craniosynostosis.

INTRODUCTION: Preoperative computed tomography (CT) evaluation of patients with nonsyndromic craniosynostosis (NSC) has focused on the bony cranial vault while ignoring the surrounding soft tissues. In this study, we posit that novel CT-derived temporal muscle and temporal fat pad morphomics (tissue thickness, area, and volume) can be used to calculate temporal morphomic indices (TMIs), which are unique to each NSC subtype (metopic, coronal, and sagittal) and divergent from normal individuals. METHODS: High-throughput image analysis was used to reconstruct the 3-dimensional anatomy and quantify a TMI. These steps were completed in a semiautomated method using algorithms programmed in MATLAB v13.0. Differences in TMI across various craniosynostosis subtypes were assessed using Wilcoxon nonparametric tests for both patients with NSC and a control cohort of patients with trauma. RESULTS: Using preoperative CT images, we evaluated 117 children with NSC from the University of Michigan Health System and 50 age-matched control patients between 1999 and 2011. Results indicate significant differences in TMI among the normal and NSC groups, with normal patients having significantly higher TMI values than patients with metopic, sagittal, and coronal synostosis. In addition, significant differences were found to exist between each craniosynostosis category. CONCLUSIONS: Patients with craniosynostosis demonstrate diminished temporalis muscle and overlying fat pad volume and thickness compared with control patients. The unique changes in temporal morphomics presented in this article demonstrate not only that the bony calvaria is affected by craniosynostosis but also that there exist quantifiable aberrations in the temporalis muscle and temporal fat pad. The methodologies described offer a novel methodology to use pre-existing CT scans to glean additional preoperative information on the soft tissue characteristics of patients with craniosynostosis.

Morphomic analysis for preoperative donor site risk assessment in patients undergoing abdominal perforator flap breast reconstruction: a proof of concept study.

BACKGROUND: Morphomics are three-dimensional measurements of aspects of the human anatomy generated by computed tomographic (CT) imaging. The purpose of this study was to generate preliminary data on the efficacy of morphomics, as a potential risk stratification tool, in predicting abdominal donor site wound-healing complications in patients undergoing abdominal perforator flap breast reconstruction. Patients and METHODS: In total, 58 consecutive patients undergoing deep inferior epigastric perforator (DIEP) flap breast reconstruction were evaluated. Using preoperative CT scan data, we quantified patients' body area, visceral and subcutaneous fat, fascia area, and body depth between T12 and L4. Associations between morphomic measures and complication rates were examined using t-tests and logistic regression. RESULTS: Of the 58 patients, 11 (19%) patients developed a wound dehiscence and 47 (81%) patients healed their abdominal incision without complications. Patients with a dehiscence had a significantly higher body mass index (BMI) (34.32 vs. 29.26 kg/m(2), p = 0.014) than patients without a dehiscence. Multiple morphometric measures including higher visceral fat area (p = 0.003) were significant predictors of abdominal donor site wound dehiscence. BMI (odds ratio [OR], 1.16; 95% confidence interval [CI], 1.03-1.32; p = 0.017) and visceral fat area (OR, 1.24; 95% CI, 1.08-1.42; p = 0.002) were independently significant predictors for wound dehiscence in the entire sample. Only visceral fat area retained its predictive ability in patients with a BMI > 30 kg/m(2). CONCLUSIONS: Morphomic measurements correlate with the likelihood of developing postoperative donor site dehiscence after DIEP flap breast reconstruction. As a proof of concept study, this demonstrates that objective data obtained from CT scans may help in preoperatively assessing the risk for donor site wound healing complications in patients undergoing DIEP flap breast reconstruction.

Machine learning methods in automated detection of CT enterography findings in Crohn's disease: A feasibility study.

PURPOSE: Qualitative findings in Crohn's disease (CD) can be challenging to reliably report and quantify. We evaluated machine learning methodologies to both standardize the detection of common qualitative findings of ileal CD and determine finding spatial localization on CT enterography (CTE). MATERIALS AND METHODS: Subjects with ileal CD and a CTE from a single center retrospective study between 2016 and 2021 were included. 165 CTEs were reviewed by two fellowship-trained abdominal radiologists for the presence and spatial distribution of five qualitative CD findings: mural enhancement, mural stratification, stenosis, wall thickening, and mesenteric fat stranding. A Random Forest (RF) ensemble model using automatically extracted specialist-directed bowel features and an unbiased convolutional neural network (CNN) were developed to predict the presence of qualitative findings. Model performance was assessed using area under the curve (AUC), sensitivity, specificity, accuracy, and kappa agreement statistics. RESULTS: In 165 subjects with 29,895 individual qualitative finding assessments, agreement between radiologists for localization was good to very good (κ = 0.66 to 0.73), except for mesenteric fat stranding (κ = 0.47). RF prediction models had excellent performance, with an overall AUC, sensitivity, specificity of 0.91, 0.81 and 0.85, respectively. RF model and radiologist agreement for localization of CD findings approximated agreement between radiologists (κ = 0.67 to 0.76). Unbiased CNN models without benefit of disease knowledge had very similar performance to RF models which used specialist-defined imaging features. CONCLUSION: Machine learning techniques for CTE image analysis can identify the presence, location, and distribution of qualitative CD findings with similar performance to experienced radiologists.

Local Assessment and Small Bowel Crohn's Disease Severity Scoring using AI.

RATIONALE AND OBJECTIVES: We present a machine learning and computer vision approach for a localized, automated, and standardized scoring of Crohn's disease (CD) severity in the small bowel, overcoming the current limitations of manual measurements CT enterography (CTE) imaging and qualitative assessments, while also considering the complex anatomy and distribution of the disease. MATERIALS AND METHODS: Two radiologists introduced a severity score and evaluated disease severity at 7.5 mm intervals along the curved planar reconstruction of the distal and terminal ileum using 236 CTE scans. A hybrid model, combining deep-learning, 3-D CNN, and Random Forest model, was developed to classify disease severity at each mini-segment. Precision, sensitivity, weighted Cohen's score, and accuracy were evaluated on a 20% hold-out test set. RESULTS: The hybrid model achieved precision and sensitivity ranging from 42.4% to 84.1% for various severity categories (normal, mild, moderate, and severe) on the test set. The model's Cohen's score (κ = 0.83) and accuracy (70.7%) were comparable to the inter-observer agreement between experienced radiologists (κ = 0.87, accuracy = 76.3%). The model accurately predicted disease length, correlated with radiologist-reported disease length (r = 0.83), and accurately identified the portion of total ileum containing moderate-to-severe disease with an accuracy of 91.51%. CONCLUSION: The proposed automated hybrid model offers a standardized, reproducible, and quantitative local assessment of small bowel CD severity and demonstrates its value in CD severity assessment.

Novel application of human morphomics to quantify temporal soft tissues in Pierre Robin and Treacher Collins.

INTRODUCTION: Pierre Robin sequence (PR) and Treacher Collins syndrome (TC) are congenital disorders associated with multiple craniofacial abnormalities. The mandibular malformations linked with these maladies are closely associated with the form and function of the temporalis muscle. Despite these associations, a paucity of research has been directed at quantifying how these malformations affect the tissues of the temporal region. In this paper, we seek to quantify differences in the temporalis muscle and the temporal fat pad using a novel CT-derived analytic program to examine craniofacial morphomic indices within these patient groups in comparison to normal age-matched controls. We posit that the temporalis muscle and temporal fat pad, like other derivatives of the first branchial arch, are hypoplastic in patients with TC and PR compared to age-matched controls. METHODS: High-throughput image analysis was used to reconstruct the 3-dimensional (3D) anatomy and quantify morphomic measures of the temporalis muscle and temporal fat pad in children with PR, TC, and age-matched controls. These steps were completed in a semi-automated method using algorithms programmed in MATLAB v13.0. The 3D reconstructions were analyzed in 3 children with PR (6 temporal regions), 3 children with TC (6 temporal regions), and a control group of 19 children (38 temporal regions). We also quantified the same measurements in a localized "core" sample in the area of greatest thickness, providing a more consistent sample of the tissue position. Relationships between the temporal muscle and fat pad values and craniofacial abnormality type were assessed using Wilcoxon nonparametric test using exact distribution, with a P value of less than 0.05 being deemed significant. RESULTS: The mean age of our patients was 6.0 years in PR and 4.5 years in TC cohorts. We were able to establish an automated methodology to quantify the temporalis muscle and temporal fat pad based on CT characteristics. Localized temporalis volume and localized temporalis area were significantly smaller in children with PR than in the control group. Total temporalis fat volume and localized temporalis area were significantly less in children with TC than in the control group. When compared to each other, the PR group had small morphomic values compared to TC group. CONCLUSIONS: There are significant morphomic differences in the temporalis muscle and the temporal fat pad in children with either PR or TC when compared to age-matched control group which can be measured from pre-existing CT scans. Specifically, both of these test groups show decreases in the morphomic measures of the temporalis region. The quantification of these changes corroborates and objectifies the clinical findings associated with these congenital deformities while simultaneously allowing for preoperative planning. Furthermore, this finding confirms that the hypoplasia seen in these patient populations is not only hypoplasia of the mandible but also of the surrounding functional matrix, which includes the temporalis muscle and temporal fat pad.

Novel temporalis muscle and fat pad morphomic analyses aids preoperative risk evaluation and outcome assessment in nonsyndromic craniosynostosis.

INTRODUCTION: Analytical morphomics is the term we created to describe an innovative, highly automated, anatomically indexed processing of 3D medical imaging data captured during the course of a patients' preoperative CT scan. Our specific aim is to determine the efficacy of craniofacial morphomic indices (CMI) such as temporalis muscle and temporal fat pad morphometric values to predict blood transfusion requirement and hospital stay in a cohort of children with nonsyndromic craniosynostosis (NSC). METHODS: High-throughput, semi-automated image analysis was used to reconstruct the 3-dimensional anatomy of the temporalis muscle and temporal fat pad and to quantify CMIs. The prognostic effect of CMI on clinical outcomes were evaluated among all NSC patients and compared across various craniosynostosis subtypes using Wilcoxon nonparametric tests and Kendall's τ to determine significance. RESULTS: Using preoperative CT images, we evaluated 117 children with NSC from the University of Michigan Health System. Results demonstrate that increased temporal fat pad volume and local temporalis muscle volume are associated with better clinical outcomes in craniosynostosis patients. More specifically, temporal fat pad volume was shown to be a significant predictor of perioperative blood transfusion requirements (P = 0.0033) and increased temporal muscle volume correlated with decreased hospital stay (P = 0.016) when controlling for other covariates including age, sex, weight, and preoperative hematocrit. In addition, the same significant predictors were found when examining individual subtypes of craniosynostosis. CONCLUSION: Our findings demonstrate that maxillofacial CT scans provide a useful quantitative index reflecting general patient health, risk stratification, and probabilities of intervention in addition to their previously established ability to determine the specific pathology of the patient. We demonstrate that temporal morphomics predict the incidence of blood transfusion, hospital stay, and serve as a proxy for fitness in patients undergoing craniosynostosis surgery.

Morphomic measurement of the temporalis muscle and zygomatic bone as novel predictors of hospital-based clinical outcomes in patients with mandible fracture.

INTRODUCTION: Patients with mandibular fracture often have comorbidities and concomitant injuries making the decision for when and how to operate a challenge. Physicians describe "temporalis wasting" as a finding that indicates frailty; however, this is a subjective finding without quantitative values. In this study, we demonstrate that decreased morphomic values of the temporalis muscle and zygomatic bone are an objective measure of frailty associated with increased injury-induced morbidity as well as negative impact on overall hospital-based clinical outcomes in patients with mandible fracture. METHODS: Computed tomographic (CT) scans from all patients with a diagnosis of a mandible fracture in the University of Michigan trauma registry and with a hospital admission were collected from the years 2004 to 2011. Automated, high-throughput CT analysis was used to reconstruct the anatomy and quantify morphomic values (temporalis volume, area and thickness, and zygomatic thickness) in these patients using MATLAB v13.0 (MathWorks Inc, Natick, MA, USA). Subsequently, a subset of 16 individuals with a Glasgow Coma Scale of 14 or 15 was analyzed to control for brain injury. Clinical data were obtained, and the association between morphomic measurements and clinical outcomes was evaluated using Pearson correlation for unadjusted analysis and multiple regression for adjusted analysis. RESULTS: The mean age of patients in the study was 47.1 years. Unadjusted analysis using Pearson correlation revealed that decreases in zygomatic bone thickness correlated strongly with increases in hospital, intensive care unit, and ventilator days (P < 0.0001, P = 0.0003, and P = 0.0017, respectively). Furthermore, we found that decreases in temporalis mean thickness correlated with increases in hospital and ventilator days (P = 0.0264 and P = 0.0306, respectively). Similarly, decreases in temporalis local mean thickness are significantly correlated with increases in hospital and ventilator days (P = 0.0232 and P = 0.0472, respectively). CONCLUSIONS: Decreased thicknesses of the zygomatic bone and temporalis muscle are significantly correlated with higher hospital, ventilator, and intensive care unit days in patients with mandibular fracture receiving reconstructive operations. This morphomic methodology provides an accurate, quantitative means to evaluate craniofacial trauma patient frailty, injury, and outcomes using routinely obtained CT scans. In the future, we plan to apply this approach to determine preoperative risk stratification and assist in surgical planning.

Using Artificial Intelligence to Predict Cirrhosis From Computed Tomography Scans.

INTRODUCTION: Undiagnosed cirrhosis remains a significant problem. In this study, we developed and tested an automated liver segmentation tool to predict the presence of cirrhosis in a population of patients with paired liver biopsy and computed tomography (CT) scans. METHODS: We used a cohort of 1,590 CT scans within the Morphomics database to train an automated liver segmentation model using 3D-U-Net and Google's DeeplLabv3+. Imaging features were then automatically calculated from an external test cohort of patients with chronic liver disease who had a paired liver biopsy and CT within 6 months of each other in January 2004-2012. Using gradient boosting decision trees, we developed multivariate models to predict the presence of histologic cirrhosis and evaluated with 5-fold cross-validated c-statistic. RESULTS: Our cohort had 351 patients; 96 patients had cirrhosis. Of the total cohort, 72 were postliver transplant. Both fibrosis (FIB)-4 and liver morphomics alone performed equally well with area under the receiving operating characteristics of 0.76 (95% confidence interval 0.70-0.81) and 0.71 (95% confidence interval 0.65-0.76), respectively ( P = 0.2). However, the combination of liver morphomics with laboratory values or liver morphomics with laboratory and demographic data resulted in significant improved performance with area under the receiving operating characteristics of 0.84 (0.80-0.89) and 0.85 (0.81-0.90), respectively, compared with FIB-4 alone ( P < 0.001). In a subgroup analysis, we also examined performance in patients without liver transplantation and saw similar augmentation of FIB-4. DISCUSSION: This proof-of-principle study demonstrates that automatically extracted features within CT scans can be combined with classic electronic medical record data to improve the prediction of cirrhosis in patients with liver disease. This tool may be used in both pretransplant and posttransplant patients and has the potential to improve our ability to detect undiagnosed cirrhosis.

Analytic Morphomics Are Related to Outcomes After Lung Volume Reduction Surgery.

:Lung volume reduction surgery continues to have a high morbidity despite National Emphysema Treatment Trial selection criteria. This study evaluated the association between analytic morphomics on chest computed tomography scans and outcomes after lung volume reduction surgery. In a retrospective review of 85 lung volume reduction surgery patients from 1998-2013, dorsal muscle group area, subcutaneous and visceral fat area, and bone mineral density were assessed using analytic morphomics. Lung density was divided into five levels of increasing density (Lung density 1, emphysema; 2, normal lung; 4-5, scarring). Outcomes including survival, hospital length of stay, readmission at 30 days, and pulmonary complications were analyzed using univariate and multivariable techniques. Pulmonary complications developed in 27.1% (23/85). Mortality at 90 days was 9.4% (8/85). On multivariable analysis, lower bone mineral density (Odds ratio 0.61; 95% confidence interval 0.39-0.95) was associated with decreased survival, longer length of stay (0.83; 0.77-0.89), and readmissions (0.39; 0.15-1.00). Higher lung density 5:lung density 2 volume (1.84; 1.05-3.23), possibly due to scarring, was associated with pulmonary complications and longer length of stay (1.32; 1.23-1.41) while lower subcutaneous fat area:height was associated with readmissions which may reflect decreased metabolic reserve (0.35; 0.13-0.93). Patients with signs of frailty including lower bone mineral density may be at increased risk of adverse outcomes including decreased survival after lung volume reduction surgery. The results of this hypothesis-generating study will need to be confirmed in larger, multicenter trials to determine whether analytic morphomics can improve risk stratification and patient selection.

Development of a quantitative method for the diagnosis of cirrhosis.

OBJECTIVE: To develop a novel non-invasive, quantitative approach utilizing computed tomography scans to predict cirrhosis. MATERIALS AND METHODS: A total of 105 patients (54 cirrhosis and 51 normal) who had CT scans within 6 months of a liver biopsy or were identified through a Trauma registry were included in this study. Patients were randomly divided into the training set (n = 81) and the validation set (n = 24). Each liver was segmented in a semi-automated fashion from a computed tomography scan using Mimics software. The resulting liver surfaces were saved as a stereo lithography mesh into an Oracle database, and analyzed in MATLAB(®) for morphological markers of cirrhosis. RESULTS: The best predictive model for diagnosing cirrhosis consisted of liver slice-bounding box slice ratio, the dimensions of the liver bounding box, liver slice area, slice perimeter, surface volume and adjusted surface area. With this model, we calculated an area under the receiver operating characteristic curve of 0.90 for the training set, and 0.91 for the validation set. For comparison, we calculated an area under the receiver operating characteristic curve of 0.70 for our dataset when we used the lab value based aspartate aminotransferase-platelet ratio index, another reported model for predicting cirrhosis. Last, by combining the aspartate aminotransferase-platelet ratio index and our model, we obtained an area under the receiving operating characteristic of 0.95. CONCLUSION: This study shows "proof of concept" that quantitative image analysis of livers on computed tomography scans may be utilized to predict cirrhosis in the absence of a liver biopsy.

The Use of Readily Available Longitudinal Data to Predict the Likelihood of Surgery in Crohn Disease.

BACKGROUND: Although imaging, endoscopy, and inflammatory biomarkers are associated with future Crohn disease (CD) outcomes, common laboratory studies may also provide prognostic opportunities. We evaluated machine learning models incorporating routinely collected laboratory studies to predict surgical outcomes in U.S. Veterans with CD. METHODS: Adults with CD from a Veterans Health Administration, Veterans Integrated Service Networks (VISN) 10 cohort examined between 2001 and 2015 were used for analysis. Patient demographics, medication use, and longitudinal laboratory values were used to model future surgical outcomes within 1 year. Specifically, data at the time of prediction combined with historical laboratory data characteristics, described as slope, distribution statistics, fluctuation, and linear trend of laboratory values, were considered and principal component analysis transformations were performed to reduce the dimensionality. Lasso regularized logistic regression was used to select features and construct prediction models, with performance assessed by area under the receiver operating characteristic using 10-fold cross-validation. RESULTS: We included 4950 observations from 2809 unique patients, among whom 256 had surgery, for modeling. Our optimized model achieved a mean area under the receiver operating characteristic of 0.78 (SD, 0.002). Anti-tumor necrosis factor use was associated with a lower probability of surgery within 1 year and was the most influential predictor in the model, and corticosteroid use was associated with a higher probability of surgery. Among the laboratory variables, high platelet counts, high mean cell hemoglobin concentrations, low albumin levels, and low blood urea nitrogen values were identified as having an elevated influence and association with future surgery. CONCLUSIONS: Using machine learning methods that incorporate current and historical data can predict the future risk of CD surgery.

Automated Measurements of Body Composition in Abdominal CT Scans Using Artificial Intelligence Can Predict Mortality in Patients With Cirrhosis.

Body composition measures derived from already available electronic medical records (computed tomography [CT] scans) can have significant value, but automation of measurements is needed for clinical implementation. We sought to use artificial intelligence to develop an automated method to measure body composition and test the algorithm on a clinical cohort to predict mortality. We constructed a deep learning algorithm using Google's DeepLabv3+ on a cohort of de-identified CT scans (n = 12,067). To test for the accuracy and clinical usefulness of the algorithm, we used a unique cohort of prospectively followed patients with cirrhosis (n = 238) who had CT scans performed. To assess model performance, we used the confusion matrix and calculated the mean accuracy of 0.977 ± 0.02 (0.975 ± 0.018 for the training and test sets, respectively). To assess for spatial overlap, we measured the mean intersection over union and mean boundary contour scores and found excellent overlap between the manual and automated methods with mean scores of 0.954 ± 0.030, 0.987 ± 0.009, and 0.948 ± 0.039 (0.983 ± 0.013 for the training and test set, respectively). Using these automated measurements, we found that body composition features were predictive of mortality in patients with cirrhosis. On multivariate analysis, the addition of body composition measures significantly improved prediction of mortality for patients with cirrhosis over Model for End-Stage Liver Disease alone (P < 0.001). Conclusion: The measurement of body composition can be automated using artificial intelligence and add significant value for incidental CTs performed for other clinical indications. This is proof of concept that this methodology could allow for wider implementation into the clinical arena.

Morphomic Factors Associated With Complete Response to Neoadjuvant Therapy in Esophageal Carcinoma.

BACKGROUND: In patients undergoing neoadjuvant chemoradiotherapy (nCRT) followed by surgery for locally advanced esophageal squamous cell carcinoma (ESCC), patients with a pathologic complete response (pCR) have the greatest benefit. The purpose of this study was to identify morphomic factors obtained from pretreatment computed tomography scans associated with a pCR in ESCC. METHODS: We retrospectively analyzed patients with ESCC treated with nCRT who underwent esophagectomy between 2006 and 2016. Clinical and morphomic characteristics pre-nCRT were analyzed to identify factors associated with pCR using univariate and multivariable analyses. RESULTS: There were 183 patients with ESCC included in this study, and 45 (24.6%) patients achieved pCR. The overall survival in patients with pCR was better than that in patients without pCR (5.8 years vs 1.2 years; P < .001). On univariate analysis, increased age, radiation dose greater than or equal to 4000 cGy, and larger subcutaneous adipose tissue area were correlated with pCR. On multivariable logistic regression, increased age (odds ratio, 1.53; P = .03), radiation dose greater than or equal to 4000 cGy (odds ratio, 2.19; P = .04), and larger dorsal muscle group normal-density area (odds ratio, 1.59; P = .03) were independently associated with pCR. CONCLUSIONS: Increased age, radiation dose greater than or equal to 4000 cGy, and larger dorsal muscle group normal-density area were significantly associated with pCR. These factors may be useful in determining which patients are most likely to benefit from nCRT followed by esophagectomy.

Assessing Small Bowel Stricturing and Morphology in Crohn's Disease Using Semi-automated Image Analysis.

BACKGROUND: Evaluating structural damage using imaging is essential for the evaluation of small intestinal Crohn's disease (CD), but it is limited by potential interobserver variation. We compared the agreement of enterography-based bowel damage measurements collected by experienced radiologists and a semi-automated image analysis system. METHODS: Patients with small bowel CD undergoing a CT-enterography (CTE) between 2011 and 2017 in a tertiary care setting were retrospectively reviewed. CT-enterography studies were reviewed by 2 experienced radiologists and separately underwent automated computer image analysis using bowel measurement software. Measurements included maximum bowel wall thickness (BWT-max), maximum bowel dilation (DIL-max), minimum lumen diameter (LUM-min), and the presence of a stricture. Measurement correlation coefficients and paired t tests were used to compare individual operator measurements. Multivariate regression was used to model identification of strictures using semi-automated measures. RESULTS: In 138 studies, the correlation between radiologists and semi-automated measures were similar for BWT-max (r = 0.724, 0.702), DIL-max (r = 0.812, 0.748), and LUM-min (r = 0.428, 0.381), respectively. Mean absolute measurement difference between semi-automated and radiologist measures were no different from the mean difference between paired radiologists for BWT-max (1.26 mm vs 1.12 mm, P = 0.857), DIL-max (2.78 mm vs 2.67 mm, P = 0.557), and LUM-min (0.54 mm vs 0.41 mm, P = 0.596). Finally, models of radiologist-defined intestinal strictures using automatically acquired measurements had an accuracy of 87.6%. CONCLUSION: Structural bowel damage measurements collected by semi-automated approaches are comparable to those of experienced radiologists. Radiomic measures of CD will become an important new data source powering clinical decision-making, patient-phenotyping, and assisting radiologists in reporting objective measures of disease status.

Analytic Morphomics Predict Outcomes After Lung Transplantation.

BACKGROUND: The purpose of this study was to identify morphomic factors on standard, pretransplantation computed tomography (CT) scans associated with outcomes after lung transplantation. METHODS: A retrospective review of 200 patients undergoing lung transplantation at a single institution from 2003 to 2014 was performed. CT scans obtained within 1 year before transplantation underwent morphomic analysis. Morphomic characteristics included lung, dorsal muscle group, bone, and subcutaneous and visceral fat area and density. Patient data were gathered from institutional and United Network for Organ Sharing databases. Outcomes, including initial ventilator support greater than 48 hours, length of stay, and survival, were evaluated using univariate and multivariable analyses. RESULTS: On multivariable Cox regression, subcutaneous fat/total body area (hazard ratio [HR] 0.60, p = 0.001), lung density 3 volume (HR 0.67, p = 0.013), and creatinine (HR 4.37, p = 0.010) were independent predictors of survival. Initial ventilator support more than 48 hours was associated with decreased vertebral body to linea alba distance (odds ratio [OR] 0.49, p = 0.002) and Zubrod score 4 (OR 14.0, p < 0.001). Increased bone mineral density (p < 0.001) and increased cross-sectional body area (p < 0.001) were associated with decreased length of stay, whereas supplemental oxygen (p < 0.001), bilateral transplantation (p = 0.002), cardiopulmonary bypass (p < 0.001), and Zubrod score 3 (p < 0.001) or 4 (p = 0.040) were associated with increased length of stay. CONCLUSIONS: Morphomic factors associated with lower metabolic reserve and frailty, including decreased subcutaneous fat, bone density, and body dimensions were independent predictors of survival, prolonged ventilation, and increased length of stay. Analytic morphomics using pretransplantation CT scans may improve recipient selection and risk stratification.