Citations

Abstract Background Gram stain (GS) is essential for preliminary bacterial identification. The lack of a standardized quality control method for GS is a significant drawback. By utilizing a deep convolutional neural network (CNN) design, we present a proof-of-concept for a novel solution to GS quality assurance. Using this approach, current manual GS, and potentially automated GS methods, may possess a standardized method of quality control. Methods Twenty-nine GS from positive blood cultures encompassing the spectrum of GS quality (acceptable, over/under decolorization, smear thickness, stain artifact) were selected by two expert microbiologists. A total of 197 images were obtained using a Cognex D-900 camera and Olympus BX-43 microscope at 1000X magnification. Images were classified using deep learning software (Cognex, ViDi Suite 4.1). Eighty-one images were classified as suboptimal and 116 were classified as optimal by expert review. To address background variation due to manual GS, we implemented a screening classification tool to separate images into 5 distinct classes (pale, rose, orange, indigo, variegated). The training and test sets consisted of 100 and 97 images respectively. Using a chaining technique, a second classification tool was added after the screening classification tool. The second tool further divided the 197 images into two dataset classes: optimal and suboptimal. The second tool used 105 images for training and 92 images for the test set. Model performance was compared to expert microbiology review, with GS verified by organism identification using MALDI-TOF. Results Cognex ViDi Suite 4.1 could discriminate between optimal and suboptimal GS images. The chained classification tool was able to be trained with 100% recall and 100% accuracy, with a resulting F-score of 1.00. Figure 1 VisionPro GUI with optimal image (MRSA) Conclusion Deep learning enabled image analysis is an effective method for sorting GS which are acceptable for further interpretation and suboptimal smears which require rework. Additional study is warranted to develop deep learning image analysis tools for the interpretation of GS on multiple specimen types and incorporate into automated GS systems. Disclosures Jane Hata, PhD , D(ABMM), Roche Molecular Systems: Grant/Research Support Kyle Magee, MLS(ASCP), SeluxDx: Employee

Objective: Parents frequently purchase and inquire about smartwatch devices to monitor child behaviors and functioning. This pilot study examined the feasibility and accuracy of using smartwatch monitoring for the prediction of disruptive behaviors. Methods: The study enrolled children ( N = 10) aged 7–10 years hospitalized for the treatment of disruptive behaviors. The study team completed continuous behavioral phenotyping during study participation. The machine learning protocol examined severe behavioral outbursts (operationalized as episodes that preceded physical restraint) for preparing the training data. Supervised machine learning methods were trained with cross-validation to predict three behavior states—calm, playful, and disruptive. Results: The participants had a 90% adherence rate for per protocol smartwatch use. Decision trees derived conditional dependencies of heart rate, sleep, and motor activity to predict behavior. A cross-validation demonstrated 80.89% accuracy of predicting the child's behavior state using these conditional dependencies. Conclusion: This study demonstrated the feasibility of 7-day continuous smartwatch monitoring for children with severe disruptive behaviors. A machine learning approach characterized predictive biomarkers of impending disruptive behaviors. Future validation studies will examine smartwatch physiological biomarkers to enhance behavioral interventions, increase parental engagement in treatment, and demonstrate target engagement in clinical trials of pharmacological agents for young children.

Abstract Methylation profiling has revealed unexpected relationships between traditional histopathology-based versus biologically-based classifications of brain tumors. Methylation class pleomorphic xanthoastrocytoma (mcPXA) comprises tumors with the same methylation signature as classical PXA but with additional morphologic patterns, including overlap with glioblastoma (GBM). To characterize mcPXA, we examined epidemiologic, histologic, and molecular features as well as clinical behavior of a cohort (n=313) of tumors matching to mcPXA by the DKFZ classifier (v12 score >= 0.85). Patient median age was 19 (range 1-73) with a female predominance (178 female/135 male). CDKN2A/B homozygous deletion was observed in 273 (87%) cases. In the subset of cases tested for BRAF p.V600E (n=120), 111 (93%) harbored the mutation. Concomitant chr7+/chr10- was observed in 66 (21%) cases. In the subset of cases tested for TERT promoter mutations (n=50), 11 (22%) harbored the mutation, always with co-occurrence of BRAF p.V600E (11/11) and infrequently with chr7+/chr10- (3/11). MGMT promoter methylation was observed in 40 (13%) cases. Overall and progression-free survival (OS, PFS) in mcPXA were similar to grade 2-3 IDH-mutant astrocytoma (A-IDH). Prognostic factors in mcPXA included GBM-like histopathology (defined as cases diagnosed as GBM prior to methylation profiling), concomitant chr7+/chr10-, and MGMT promoter methylation. GBM-like histopathology was associated with shorter OS (p=0.02) and PFS (p< 0.01) compared to cases without GBM-like histopathology. Concomitant chr7+/chr10- was associated with shorter PFS (p=0.02), and MGMT promoter methylation was associated with shorter OS (p=0.03). Our findings demonstrate that mcPXA tumors affect adolescents and young adults, share molecular features of classical PXA, and manifest clinical behavior similar to classical PXA, but subsets of tumors with atypical features (chr7+/chr10-, MGMT promoter methylation, and/or GBM-like histopathology) are more aggressive. The use of a molecular definition of PXA reveals that this tumor type is more common than previously appreciated and warrants in-depth study for therapeutic options.

Abstract High-grade glioma continues to have dismal survival owing in part to its intra- and inter-patient heterogeneity. Standard clinical protocol collects tumor samples with the aim of providing or confirming a diagnosis and determining the status of a few key genes (e.g. IDH1, MGMT). However, this protocol is unable to capture the diversity within tumor regions, immune expression or normal cell abundances that play key roles in the development of the disease. To overcome this, during surgery we collect image-localized multi-regional biopsies to characterize this disease heterogeneity. Data collection is ongoing, and we currently have 202 samples from 58 patients with available bulk RNA-Seq. With a single-cell reference dataset from Columbia University, we used CIBERSORTx, a deconvolution method, to predict relative abundances of 7 normal, 6 glioma, and 5 immune cell subpopulations for each sample. We used Cox proportional hazard models with first-order statistics (mean, minimum, maximum) of abundances within patients to determine whether these are prognostic, then used TCGA RNA-Seq data to validate these findings. We found that one glioma proneural subtype was significantly beneficial for patient survival relative to other glioma subtypes across all statistics (and showed significance in TCGA). Proliferative and mesenchymal glioma subtypes also showed significance for one or two of the calculated statistics. Oligodendrocyte progenitor cell abundances were consistently significantly beneficial for patient survival in our cohort, while an increased abundance of abnormal (reactive) astrocytes is associated with poor prognosis. We also ran these analyses within invasive margin and core tumor regions to determine location-specific population abundances associated with patient survival. In conclusion, understanding the in vivo diversity of cellular subpopulations within high grade glioma is important for treatment stratification and patient benefit.

Abstract Identification of key phenotypic regions such as necrosis, contrast enhancement, and edema on magnetic resonance imaging (MRI) is important for understanding disease evolution and treatment response in patients with glioma. Manual delineation is time intensive and not feasible for a clinical workflow. Automating phenotypic region segmentation overcomes many issues with manual segmentation, however, current glioma segmentation datasets focus on pre-treatment, diagnostic scans, where treat-ment effects and surgical cavities are not present. Thus, existing automatic segmenta-tion models are not applicable to post-treatment imaging that is used for longitudinal evaluation of care. Here, we present a comparison of three-dimensional convolutional neural networks (nnU-Net architecture) trained on large temporally defined pre-treatment, post-treatment, and mixed cohorts. We used a total of 1563 imaging timepoints from 854 patients curated from 13 different institutions as well as diverse public data sets to understand the capabilities and limitations of automatic segmenta-tion on glioma images with different phenotypic and treatment appearance. We as-sessed the performance of models using Dice coefficients on test cases from each group comparing predictions with manual segmentations generated by trained techni-cians. We demonstrate that training a combined model can be as effective as models trained on just one temporal group. The results highlight the importance of a diverse training set, that includes images from the course of disease and with effects from treatment, in the creation of a model that can accurately segment glioma MRIs at multiple treatment time points.

Abstract Image-based mathematical modeling is emerging as an advantageous tool to integrate into the care routine for patients with brain tumors. These tools can be used for a variety of tasks including predicting tumor growth, indicating spatial genomic alterations, and guiding radiation therapy. Often, the methods to produce these maps are developed in silos, using cohorts of retrospective patients and switching to a per patient map calculation can be an afterthought. Automatic pipelines are needed to enable creating mathematical maps in a clinically-relevant workflow. To this end, we developed a pipeline to create an image-based mathematical map of edema abundance based on T2-weighted (T2W) magnetic resonance imaging that includes preprocessing steps, abnormality and tissue type segmentation, brain extraction, map generation, and conversion between image file formats. The python-based pipeline was designed to take in T1-weighted (T1W), T1W with gadolinium contrast (T1Gd), T2W, and fluid attenuated inversion recovery (FLAIR) DICOM images for a single patient as input and convert them to NIfTIs for downstream steps. The pipeline preprocessed images by correcting bias field fluctuations using the N4 algorithm and normalizing intensities. The T1W, T1Gd, and FLAIR are then registered to the T2W. T1Gd and FLAIR are used in a deep learning algorithm to delineate the brain and abnormality including enhancing tumor, necrosis, and peripheral edema. Normal tissue segmentations are performed using the statistical parametric mapping (SPM) segmentation routine. The segmentations and preprocessed images are all utilized in the edema mathematical model. The resulting edema map is then reformatted as a DICOM and output for further use by clinicians and clinical imaging systems. Our future goals include integrating our pipeline in a clinical workstation and sending final maps to our hospital’s picture archiving and communication system (PACS) to allow for deployment of these models for patient care in a clinically efficient timeline.

Abstract Rationale Follow-up of patients with emphysema treated with endobronchial valves is limited to 3–12 months after treatment in prior reports. To date, no comparative data exist between treatment and control subjects with a longer follow-up. Objectives To assess the durability of the Spiration Valve System (SVS) in patients with severe heterogeneous emphysema over a 24-month period. Methods EMPROVE, a multicenter randomized controlled trial, presents a rigorous comparison between treatment and control groups for up to 24 months. Lung function, respiratory symptoms, and quality-of-life (QOL) measures were assessed. Results A significant improvement in forced expiratory volume in 1 second was maintained at 24 months in the SVS treatment group versus the control group. Similarly, significant improvements were maintained in several QOL measures, including the St. George’s Respiratory Questionnaire and the COPD Assessment Test. Patients in the SVS treatment group experienced significantly less dyspnea than those in the control group, as indicated by the modified Medical Research Council dyspnea scale score. Adverse events at 24 months did not significantly differ between the SVS treatment and control groups. Acute chronic obstructive pulmonary disease exacerbation rates in the SVS treatment and control groups were 13.7% (14 of 102) and 15.6% (7 of 45), respectively. Pneumothorax rates in the SVS treatment and control groups were 1.0% (1 of 102) and 0.0% (0 of 45), respectively. Conclusions SVS treatment resulted in statistically significant and clinically meaningful durable improvements in lung function, respiratory symptoms, and QOL, as well as a statistically significant reduction in dyspnea, for at least 24 months while maintaining an acceptable safety profile. Clinical trial registered with www.clinicaltrials.gov (NCT01812447).

Abstract INTRODUCTION Magnetic resonance imaging (MRI) research has advanced our understanding of neurodegeneration in sporadic early‐onset Alzheimer's disease (EOAD) but studies include small samples, mostly amnestic EOAD, and have not focused on developing an MRI biomarker. METHODS We analyzed MRI scans to define the sporadic EOAD‐signature atrophy in a small sample (n = 25) of Massachusetts General Hospital (MGH) EOAD patients, investigated its reproducibility in the large longitudinal early‐onset Alzheimer's disease study (LEADS) sample (n = 211), and investigated the relationship of the magnitude of atrophy with cognitive impairment. RESULTS The EOAD‐signature atrophy was replicated across the two cohorts, with prominent atrophy in the caudal lateral temporal cortex, inferior parietal lobule, and posterior cingulate and precuneus cortices, and with relative sparing of the medial temporal lobe. The magnitude of EOAD‐signature atrophy was associated with the severity of cognitive impairment. DISCUSSION The EOAD‐signature atrophy is a reliable and clinically valid biomarker of AD‐related neurodegeneration that could be used in clinical trials for EOAD. Highlights We developed an early‐onset Alzheimer's disease (EOAD)–signature of atrophy based on magnetic resonance imaging (MRI) scans. EOAD signature was robustly reproducible across two independent patient cohorts. EOAD signature included prominent atrophy in parietal and posterior temporal cortex. The EOAD‐signature atrophy was associated with the severity of cognitive impairment. EOAD signature is a reliable and clinically valid biomarker of neurodegeneration.