A Method for Efficient De-identification of DICOM Metadata and Burned-in Pixel Text.

De-identification of DICOM images is an essential component of medical image research. While many established methods exist for the safe removal of protected health information (PHI) in DICOM metadata, approaches for the removal of PHI "burned-in" to image pixel data are typically manual, and automated high-throughput approaches are not well validated. Emerging optical character recognition (OCR) models can potentially detect and remove PHI-bearing text from medical images but are very time-consuming to run on the high volume of images found in typical research studies. We present a data processing method that performs metadata de-identification for all images combined with a targeted approach to only apply OCR to images with a high likelihood of burned-in text. The method was validated on a dataset of 415,182 images across ten modalities representative of the de-identification requests submitted at our institution over a 20-year span. Of the 12,578 images in this dataset with burned-in text of any kind, only 10 passed undetected with the method. OCR was only required for 6050 images (1.5% of the dataset).

Duke Liver Dataset: A Publicly Available Liver MRI Dataset with Liver Segmentation Masks and Series Labels.

The Duke Liver Dataset contains 2146 abdominal MRI series from 105 patients, including a majority with cirrhotic features, and 310 image series with corresponding manually segmented liver masks.

Systematic Analysis of Common Factors Impacting Deep Learning Model Generalizability in Liver Segmentation.

Purpose: To investigate the effect of training data type on generalizability of deep learning liver segmentation models. Materials and Methods: This Health Insurance Portability and Accountability Act-compliant retrospective study included 860 MRI and CT abdominal scans obtained between February 2013 and March 2018 and 210 volumes from public datasets. Five single-source models were trained on 100 scans each of T1-weighted fat-suppressed portal venous (dynportal), T1-weighted fat-suppressed precontrast (dynpre), proton density opposed-phase (opposed), single-shot fast spin-echo (ssfse), and T1-weighted non-fat-suppressed (t1nfs) sequence types. A sixth multisource (DeepAll) model was trained on 100 scans consisting of 20 randomly selected scans from each of the five source domains. All models were tested against 18 target domains from unseen vendors, MRI types, and modality (CT). The Dice-Sørensen coefficient (DSC) was used to quantify similarity between manual and model segmentations. Results: Single-source model performance did not degrade significantly against unseen vendor data. Models trained on T1-weighted dynamic data generally performed well on other T1-weighted dynamic data (DSC = 0.848 ± 0.183 [SD]). The opposed model generalized moderately well to all unseen MRI types (DSC = 0.703 ± 0.229). The ssfse model failed to generalize well to any other MRI type (DSC = 0.089 ± 0.153). Dynamic and opposed models generalized moderately well to CT data (DSC = 0.744 ± 0.206), whereas other single-source models performed poorly (DSC = 0.181 ± 0.192). The DeepAll model generalized well across vendor, modality, and MRI type and against externally sourced data. Conclusion: Domain shift in liver segmentation appears to be tied to variations in soft-tissue contrast and can be effectively bridged with diversification of soft-tissue representation in training data.Keywords: Convolutional Neural Network (CNN), Deep Learning Algorithms, Machine Learning Algorithms, Supervised Learning, CT, MRI, Liver Segmentation Supplemental material is available for this article. © RSNA, 2023.

Phase II Investigation of TVB-2640 (Denifanstat) with Bevacizumab in Patients with First Relapse High-Grade Astrocytoma.

PURPOSE: Glioblastoma represents the most common primary brain tumor. Although antiangiogenics are used in the recurrent setting, they do not prolong survival. Glioblastoma is known to upregulate fatty acid synthase (FASN) to facilitate lipid biosynthesis. TVB-2640, a FASN inhibitor, impairs this activity. PATIENTS AND METHODS: We conducted a prospective, single-center, open-label, unblinded, phase II study of TVB-2640 plus bevacizumab in patients with recurrent high-grade astrocytoma. Patients were randomly assigned to TVB-2640 (100 mg/m2 oral daily) plus bevacizumab (10 mg/kg i.v., D1 and D15) or bevacizumab monotherapy for cycle 1 only (28 days) for biomarker analysis. Thereafter, all patients received TVB-2640 plus bevacizumab until treatment-related toxicity or progressive disease (PD). The primary endpoint was progression-free survival (PFS). RESULTS: A total of 25 patients were enrolled. The most frequently reported adverse events (AE) were palmar-plantar erythrodysesthesia, hypertension, mucositis, dry eye, fatigue, and skin infection. Most were grade 1 or 2 in intensity. The overall response rate (ORR) for TVB-2640 plus bevacizumab was 56% (complete response, 17%; partial response, 39%). PFS6 for TVB-2640 plus bevacizumab was 31.4%. This represented a statistically significant improvement in PFS6 over historical bevacizumab monotherapy (BELOB 16%; P = 0.008) and met the primary study endpoint. The observed OS6 was 68%, with survival not reaching significance by log-rank test (P = 0.56). CONCLUSIONS: In this phase II study of relapsed high-grade astrocytoma, TVB-2640 was found to be a well-tolerated oral drug that could be safely combined with bevacizumab. The favorable safety profile and response signals support the initiation of a larger multicenter trial of TVB-2640 plus bevacizumab in astrocytoma.

Towards optimal deep fusion of imaging and clinical data via a model-based description of fusion quality.

BACKGROUND: Due to intrinsic differences in data formatting, data structure, and underlying semantic information, the integration of imaging data with clinical data can be non-trivial. Optimal integration requires robust data fusion, that is, the process of integrating multiple data sources to produce more useful information than captured by individual data sources. Here, we introduce the concept of fusion quality for deep learning problems involving imaging and clinical data. We first provide a general theoretical framework and numerical validation of our technique. To demonstrate real-world applicability, we then apply our technique to optimize the fusion of CT imaging and hepatic blood markers to estimate portal venous hypertension, which is linked to prognosis in patients with cirrhosis of the liver. PURPOSE: To develop a measurement method of optimal data fusion quality deep learning problems utilizing both imaging data and clinical data. METHODS: Our approach is based on modeling the fully connected layer (FCL) of a convolutional neural network (CNN) as a potential function, whose distribution takes the form of the classical Gibbs measure. The features of the FCL are then modeled as random variables governed by state functions, which are interpreted as the different data sources to be fused. The probability density of each source, relative to the probability density of the FCL, represents a quantitative measure of source-bias. To minimize this source-bias and optimize CNN performance, we implement a vector-growing encoding scheme called positional encoding, where low-dimensional clinical data are transcribed into a rich feature space that complements high-dimensional imaging features. We first provide a numerical validation of our approach based on simulated Gaussian processes. We then applied our approach to patient data, where we optimized the fusion of CT images with blood markers to predict portal venous hypertension in patients with cirrhosis of the liver. This patient study was based on a modified ResNet-152 model that incorporates both images and blood markers as input. These two data sources were processed in parallel, fused into a single FCL, and optimized based on our fusion quality framework. RESULTS: Numerical validation of our approach confirmed that the probability density function of a fused feature space converges to a source-specific probability density function when source data are improperly fused. Our numerical results demonstrate that this phenomenon can be quantified as a measure of fusion quality. On patient data, the fused model consisting of both imaging data and positionally encoded blood markers at the theoretically optimal fusion quality metric achieved an AUC of 0.74 and an accuracy of 0.71. This model was statistically better than the imaging-only model (AUC = 0.60; accuracy = 0.62), the blood marker-only model (AUC = 0.58; accuracy = 0.60), and a variety of purposely sub-optimized fusion models (AUC = 0.61-0.70; accuracy = 0.58-0.69). CONCLUSIONS: We introduced the concept of data fusion quality for multi-source deep learning problems involving both imaging and clinical data. We provided a theoretical framework, numerical validation, and real-world application in abdominal radiology. Our data suggests that CT imaging and hepatic blood markers provide complementary diagnostic information when appropriately fused.

CT-derived body composition measurements as predictors for neoadjuvant treatment tolerance and survival in gastroesophageal adenocarcinoma.

PURPOSE: Treatment for gastroesophageal adenocarcinomas can result in significant morbidity and mortality. The purpose of this study is to supplement methods for choosing treatment strategy by assessing the relationship between CT-derived body composition, patient, and tumor features, and clinical outcomes in this population. METHODS: Patients with neoadjuvant treatment, biopsy-proven gastroesophageal adenocarcinoma, and initial staging CTs were retrospectively identified from institutional clinic encounters between 2000 and 2019. Details about patient, disease, treatment, and outcomes (including therapy tolerance and survival) were extracted from electronic medical records. A deep learning semantic segmentation algorithm was utilized to measure cross-sectional areas of skeletal muscle (SM), visceral fat (VF), and subcutaneous fat (SF) at the L3 vertebra level on staging CTs. Univariate and multivariate analyses were performed to assess the relationships between predictors and outcomes. RESULTS: 142 patients were evaluated. Median survival was 52 months. Univariate and multivariate analysis showed significant associations between treatment tolerance and SM and VF area, SM to fat and VF to SF ratios, and skeletal muscle index (SMI) (p = 0.004-0.04). Increased survival was associated with increased body mass index (BMI) (p = 0.01) and increased SMI (p = 0.004). A multivariate Cox model consisting of BMI, SMI, age, gender, and stage demonstrated that patients in the high-risk group had significantly lower survival (HR = 1.77, 95% CI = 1.13-2.78, p = 0.008). CONCLUSION: CT-based measures of body composition in patients with gastroesophageal adenocarcinoma may be independent predictors of treatment complications and survival and can supplement methods for assessing functional status during treatment planning.

Radiomics: a primer on high-throughput image phenotyping.

Radiomics is a high-throughput approach to image phenotyping. It uses computer algorithms to extract and analyze a large number of quantitative features from radiological images. These radiomic features collectively describe unique patterns that can serve as digital fingerprints of disease. They may also capture imaging characteristics that are difficult or impossible to characterize by the human eye. The rapid development of this field is motivated by systems biology, facilitated by data analytics, and powered by artificial intelligence. Here, as part of Abdominal Radiology's special issue on Quantitative Imaging, we provide an introduction to the field of radiomics. The technique is formally introduced as an advanced application of data analytics, with illustrating examples in abdominal radiology. Artificial intelligence is then presented as the main driving force of radiomics, and common techniques are defined and briefly compared. The complete step-by-step process of radiomic phenotyping is then broken down into five key phases. Potential pitfalls of each phase are highlighted, and recommendations are provided to reduce sources of variation, non-reproducibility, and error associated with radiomics.

Missed Incidental Pulmonary Embolism: Harnessing Artificial Intelligence to Assess Prevalence and Improve Quality Improvement Opportunities.

PURPOSE: Incidental pulmonary embolism (IPE) can be found on body CT. The aim of this study was to evaluate the feasibility of using artificial intelligence to identify missed IPE on a large number of CT examinations. METHODS: This retrospective analysis included all single-phase chest, abdominal, and pelvic (CAP) and abdominal and pelvic (AP) CT examinations performed at a single center over 1 year, for indications other than identification of PE. Proprietary visual classification and natural language processing software was used to analyze images and reports from all CT examinations, followed by a two-step human adjudication process to classify cases as true positive, false positive, true negative, or false negative. Descriptive statistics were assessed for prevalence of IPE and features (subsegmental versus central, unifocal versus multifocal, right heart strain or not) of missed IPE. Interrater agreement for radiologist readers was also calculated. RESULTS: A total of 11,913 CT examinations (6,398 CAP, 5,515 AP) were included. Thirty false-negative examinations were identified on CAP (0.47%; 95% confidence interval [CI], 0.32%-0.67%) and nineteen false-negative studies on AP (0.34%; 95% CI, 0.21%-0.54%) studies. During manual review, readers showed substantial agreement for identification of IPE on CAP (κ = 0.76; 95% CI, 0.66-0.86) and nearly perfect agreement for identification of IPE on AP (κ = 0.86; 95% CI, 0.76-0.97). Forty-nine missed IPEs (0.41%; 95% CI, 0.30%-0.54%) were ultimately identified, compared with seventy-nine IPEs (0.66%; 95% CI, 0.53%-0.83%) identified at initial clinical interpretation. CONCLUSIONS: Artificial intelligence can efficiently analyze CT examinations to identify potential missed IPE. These results can inform peer-review efforts and quality control and could potentially be implemented in a prospective fashion.

Fully automated analysis of OCT imaging of human kidneys for prediction of post-transplant function.

Current measures for assessing the viability of donor kidneys are lacking. Optical coherence tomography (OCT) can image subsurface tissue morphology to supplement current measures and potentially improve prediction of post-transplant function. OCT imaging was performed on donor kidneys before and immediately after implantation during 169 human kidney transplant surgeries. A system for automated image analysis was developed to measure structural parameters of the kidney's proximal convoluted tubules (PCTs) visualized in the OCT images. The association of these structural parameters with post-transplant function was investigated. This study included kidneys from live and deceased donors. 88 deceased donor kidneys in this study were stored by static cold storage (SCS) and an additional 15 were preserved by hypothermic machine perfusion (HMP). A subset of both SCS and HMP deceased donor kidneys were classified as expanded criteria donor (ECD) kidneys, with elevated risk of poor post-transplant function. Post-transplant function was characterized as either immediate graft function (IGF) or delayed graft function (DGF). In ECD kidneys stored by SCS, increased PCT lumen diameter was found to predict DGF both prior to implantation and following reperfusion. In SCD kidneys preserved by HMP, reduced distance between adjacent lumen following reperfusion was found to predict DGF. Results suggest that OCT measurements may be useful for predicting post-transplant function in ECD kidneys and kidneys stored by HMP. OCT analysis of donor kidneys may aid in allocation of kidneys to expand the donor pool as well as help predict post-transplant function in transplanted kidneys to inform post-operative care.

Pgc-1α repression and high-fat diet induce age-related macular degeneration-like phenotypes in mice.

Age-related macular degeneration (AMD) is the major cause of blindness in the elderly in developed countries and its prevalence is increasing with the aging population. AMD initially affects the retinal pigment epithelium (RPE) and gradually leads to secondary photoreceptor degeneration. Recent studies have associated mitochondrial damage with AMD, and we have observed mitochondrial and autophagic dysfunction and repressed peroxisome proliferator-activated receptor-γ coactivator 1α (PGC-1α; also known as Ppargc1a) in native RPE from AMD donor eyes and their respective induced pluripotent stem cell-derived RPE. To further investigate the effect of PGC-1α repression, we have established a mouse model by feeding Pgc-1α+/- mice with a high-fat diet (HFD) and investigated RPE and retinal health. We show that when mice expressing lower levels of Pgc-1α are exposed to HFD, they present AMD-like abnormalities in RPE and retinal morphology and function. These abnormalities include basal laminar deposits, thickening of Bruch's membrane with drusen marker-containing deposits, RPE and photoreceptor degeneration, decreased mitochondrial activity, increased levels of reactive oxygen species, decreased autophagy dynamics/flux, and increased inflammatory response in the RPE and retina. Our study shows that Pgc-1α is important in outer retina biology and that Pgc-1α+/- mice fed with HFD provide a promising model to study AMD, opening doors for novel treatment strategies.