Sex classification of 3D skull images using deep neural networks.

Determining the fundamental characteristics that define a face as "feminine" or "masculine" has long fascinated anatomists and plastic surgeons, particularly those involved in aesthetic and gender-affirming surgery. Previous studies in this area have relied on manual measurements, comparative anatomy, and heuristic landmark-based feature extraction. In this study, we collected retrospectively at Cedars Sinai Medical Center (CSMC) a dataset of 98 skull samples, which is the first dataset of this kind of 3D medical imaging. We then evaluated the accuracy of multiple deep learning neural network architectures on sex classification with this dataset. Specifically, we evaluated methods representing three different 3D data modeling approaches: Resnet3D, PointNet++, and MeshNet. Despite the limited number of imaging samples, our testing results show that all three approaches achieve AUC scores above 0.9 after convergence. PointNet++ exhibits the highest accuracy, while MeshNet has the lowest. Our findings suggest that accuracy is not solely dependent on the sparsity of data representation but also on the architecture design, with MeshNet's lower accuracy likely due to the lack of a hierarchical structure for progressive data abstraction. Furthermore, we studied a problem related to sex determination, which is the analysis of the various morphological features that affect sex classification. We proposed and developed a new method based on morphological gradients to visualize features that influence model decision making. The method based on morphological gradients is an alternative to the standard saliency map, and the new method provides better visualization of feature importance. Our study is the first to develop and evaluate deep learning models for analyzing 3D facial skull images to identify imaging feature differences between individuals assigned male or female at birth. These findings may be useful for planning and evaluating craniofacial surgery, particularly gender-affirming procedures, such as facial feminization surgery.

Single protein encapsulated SN38 for tumor-targeting treatment.

BACKGROUND: The alkaloid camptothecin analog SN38 is a potent antineoplastic agent, but cannot be used directly for clinical application due to its poor water solubility. Currently, the prodrug approach on SN38 has resulted in 3 FDA-approved cancer therapeutics, irinotecan, ONIVYDE, and Trodelvy. However, only 2-8% of irinotecan can be transformed enzymatically in vivo into the active metabolite SN38, which severely limits the drug's efficacy. While numerous drug delivery systems have been attempted to achieve effective SN38 delivery, none have produced drug products with antitumor efficacy better than irinotecan in clinical trials. Therefore, novel approaches are urgently needed for effectively delivering SN38 to cancer cells with better efficacy and lower toxicity. METHODS: Based on the unique properties of human serum albumin (HSA), we have developed a novel single protein encapsulation (SPE) technology to formulate cancer therapeutics for improving their pharmacokinetics (PK) and antitumor efficacy and reducing their side effects. Previous application of SPE technology to doxorubicin (DOX) formulation has led to a promising drug candidate SPEDOX-6 (FDA IND #, 152154), which will undergo a human phase I clinical trial. Using the same SPE platform on SN38, we have now produced two SPESN38 complexes, SPESN38-5 and SPESN38-8. We conducted their pharmacological evaluations with respect to maximum tolerated dose, PK, and in vivo efficacy against colorectal cancer (CRC) and soft tissue sarcoma (STS) in mouse models. RESULTS: The lyophilized SPESN38 complexes can dissolve in aqueous media to form clear and stable solutions. Maximum tolerated dose (MTD) of SPESN38-5 is 250 mg/kg by oral route (PO) and 55 mg/kg by intravenous route (IV) in CD-1 mice. SPESN38-8 has the MTD of 45 mg/kg by IV in the same mouse model. PK of SPESN38-5 by PO at 250 mg/kg gave mouse plasma AUC0-∞ of 0.05 and 4.5 nmol × h/mL for SN38 and SN38 glucuronidate (SN38G), respectively, with a surprisingly high molar ratio of SN38G:SN38 = 90:1. However, PK of SPESN38-5 by IV at 55 mg/kg yielded much higher mouse plasma AUC0-∞ of 19 and 28 nmol × h/mL for SN38 and SN38G, producing a much lower molar ratio of SN38G:SN38 = 1.5:1. Antitumor efficacy of SPESN38-5 and irinotecan (control) was evaluated against HCT-116 CRC xenograft tumors. The data indicates that SPESN38-5 by IV at 55 mg/kg is more effective in suppressing HCT-116 tumor growth with lower systemic toxicity compared to irinotecan at 50 mg/kg. Additionally, SPESN38-8 and DOX (control) by IV were evaluated in the SK-LMS-1 STS mouse model. The results show that SPESN38-8 at 33 mg/kg is highly effective for inhibiting SK-LMS-1 tumor growth with low toxicity, in contrast to DOX's insensitivity to SK-LMS-1 with high toxicity. CONCLUSION: SPESN38 complexes provide a water soluble SN38 formulation. SPESN38-5 and SPESN38-8 demonstrate better PK values, lower toxicity, and superior antitumor efficacy in mouse models, compared with irinotecan and DOX.

Single Protein Encapsulated SN38 for Tumor-Targeting Treatment.

Background: The alkaloid camptothecin analog SN38 is a potent antineoplastic agent, but cannot be used directly for clinical application due to its poor water solubility. Currently, the prodrug approach on SN38 has resulted in 3 FDA-approved cancer therapeutics, irinotecan, ONIVYDE, and Trodelvy. However, only 2-8% of irinotecan can be transformed enzymatically in vivo into the active metabolite SN38, which severely limits the drug's efficacy. While numerous drug delivery systems have been attempted to achieve effective SN38 delivery, none have produced drug products with antitumor efficacy better than irinotecan in clinical trials. Therefore, novel approaches are urgently needed for effectively delivering SN38 to cancer cells with better efficacy and lower toxicity. Methods: Based on the unique properties of human serum albumin (HSA), we have developed a novel single protein encapsulation (SPE) technology to formulate cancer therapeutics for improving their pharmacokinetics (PK) and antitumor efficacy and reducing their side effects. Previous application of SPE technology to doxorubicin (DOX) formulation has led to a promising drug candidate SPEDOX-6 (FDA IND #, 152154), which will undergo a human phase I clinical trial. Using the same SPE platform on SN38, we have now produced two SPESN38 complexes, SPESN38-5 and SPESN38-8. We conducted their pharmacological evaluations with respect to maximum tolerated dose, PK, and in vivo efficacy against colorectal cancer (CRC) and soft tissue sarcoma (STS) in mouse models. Results: The lyophilized SPESN38 complexes can dissolve in aqueous media to form clear and stable solutions. Maximum tolerated dose (MTD) of SPESN38-5 is 250 mg/kg by oral route (PO) and 55 mg/kg by intravenous route (IV) in CD-1 mice. SPESN38-8 has the MTD of 45 mg/kg by IV in the same mouse model. PK of SPESN38-5 by PO at 250 mg/kg gave mouse plasma AUC0-∞ of 0.0548 and 4.5007 (nmol × h/mL) for SN38 and SN38 glucuronidate (SN38G), respectively, with a surprisingly high molar ratio of SN38G:SN38 = 82:1. However, PK of SPESN38-5 by IV at 55 mg/kg yielded much higher mouse plasma AUC0-∞ of 18.80 and 27.78 nmol × h/mL for SN38 and SN38G, producing a much lower molar ratio of SN38G:SN38 = 1.48:1. Antitumor efficacy of SPESN38-5 and irinotecan (control) was evaluated against HCT-116 CRC xenograft tumors. The data indicates that SPESN38-5 by IV at 55 mg/kg is more effective in suppressing HCT-116 tumor growth with lower systemic toxicity compared to irinotecan at 50 mg/kg. Additionally, SPESN38-8 and DOX (control) by IV were evaluated in the SK-LMS-1 STS mouse model. The results show that SPESN38-8 at 33 mg/kg is highly effective for inhibiting SK-LMS-1 tumor growth with low toxicity, in contrast to DOX's insensitivity to SK-LMS-1 with high toxicity. Conclusion: SPESN38 complexes provide a water soluble SN38 formulation. SPESN38-5 and SPESN38-8 demonstrate better PK values, lower toxicity, and superior antitumor efficacy in mouse models, compared with irinotecan and DOX.

COVID19 Diagnosis Using Chest X-rays and Transfer Learning.

A pandemic of respiratory illnesses from a novel coronavirus known as Sars-CoV-2 has swept across the globe since December of 2019. This is calling upon the research community including medical imaging to provide effective tools for use in combating this virus. Research in biomedical imaging of viral patients is already very active with machine learning models being created for diagnosing Sars-CoV-2 infections in patients using CT scans and chest x-rays. We aim to build upon this research. Here we used a transfer-learning approach to develop models capable of diagnosing COVID19 from chest x-ray. For this work we compiled a dataset of 112120 negative images from the Chest X-Ray 14 and 2725 positive images from public repositories. We tested multiple models, including logistic regression and random forest and XGBoost with and without principal components analysis, using five-fold cross-validation to evaluate recall, precision, and f1-score. These models were compared to a pre-trained deep-learning model for evaluating chest x-rays called COVID-Net. Our best model was XGBoost with principal components with a recall, precision, and f1-score of 0.692, 0.960, 0.804 respectively. This model greatly outperformed COVID-Net which scored 0.987, 0.025, 0.048. This model, with its high precision and reasonable sensitivity, would be most useful as "rule-in" test for COVID19. Though it outperforms some chemical assays in sensitivity, this model should be studied in patients who would not ordinarily receive a chest x-ray before being used for screening.

Detecting Intra-Field Variation in Rice Yield With Unmanned Aerial Vehicle Imagery and Deep Learning.

Unmanned aerial vehicles (UAVs) equipped with multispectral sensors offer high spatial and temporal resolution imagery for monitoring crop stress at early stages of development. Analysis of UAV-derived data with advanced machine learning models could improve real-time management in agricultural systems, but guidance for this integration is currently limited. Here we compare two deep learning-based strategies for early warning detection of crop stress, using multitemporal imagery throughout the growing season to predict field-scale yield in irrigated rice in eastern Arkansas. Both deep learning strategies showed improvements upon traditional statistical learning approaches including linear regression and gradient boosted decision trees. First, we explicitly accounted for variation across developmental stages using a 3D convolutional neural network (CNN) architecture that captures both spatial and temporal dimensions of UAV images from multiple time points throughout one growing season. 3D-CNNs achieved low prediction error on the test set, with a Root Mean Squared Error (RMSE) of 8.8% of the mean yield. For the second strategy, a 2D-CNN, we considered only spatial relationships among pixels for image features acquired during a single flyover. 2D-CNNs trained on images from a single day were most accurate when images were taken during booting stage or later, with RMSE ranging from 7.4 to 8.2% of the mean yield. A primary benefit of convolutional autoencoder-like models (based on analyses of prediction maps and feature importance) is the spatial denoising effect that corrects yield predictions for individual pixels based on the values of vegetation index and thermal features for nearby pixels. Our results highlight the promise of convolutional autoencoders for UAV-based yield prediction in rice.

Study COVID-19 Severity of Patients Admitted to Emergency Room (ER) with Chest X-ray Images.

We have conducted a study of the COVID-19 severity with the chest x-ray images, a private dataset collected from our collaborator St Bernards Medical Center. The dataset is comprised of chest x-ray images from 1,550 patients who were admitted to emergency room (ER) and were all tested positive for COVID-19. Our study is focused on the following two questions: (1) To predict patients hospital staying duration, based on the chest x-ray image which was taken when the patient was admitted to the ER. The length of stay ranged from zero hours to 95 days in the hospital and followed a power law distribution. Based on our testing results, it is hard for the prediction models to detect strong signal from the chest x-ray images. No model was able to perform better than a trivial most-frequent classifier. However, each model was able to outperform the most-frequent classifier when the data was split evenly into four categories. This would suggest that there is signal in the images, and the performance may be further improved by the addition of clinical features as well as increasing the training set. (2) To predict if a patient is COVID-19 positive or not with the chest x-ray image. We also tested the generalizability of training a prediction model on chest x-ray images from one hospital and then testing the model on images captures from other sites. With our private dataset and the COVIDx dataset, the prediction model can achieve a high accuracy of 95.9%. However, for our hold-one-out study of the generalizability of the models trained on chest x-rays, we found that the model performance suffers due to a significant reduction in training samples of any class.

An Ensemble of U-Net Models for Kidney Tumor Segmentation With CT Images.

We present here the Arkansas AI-Campus solution method for the 2019 Kidney Tumor Segmentation Challenge (KiTS19). Our Arkansas AI-Campus team participated the KiTS19 Challenge for four months, from March to July of 2019. This paper provides a summary of our methods, training, testing and validation results for this grand challenge in biomedical imaging analysis. Our deep learning model is an ensemble of U-Net models developed after testing many model variations. Our model has consistent performance on the local test dataset and the final competition independent test dataset. The model achieved local test Dice scores of 0.949 for kidney and tumor segmentation, and 0.601 for tumor segmentation, and the final competition test earned Dice scores 0.9470 and 0.6099 respectively. The Arkansas AI-Campus team solution with a composite DICE score of 0.7784 has achieved a final ranking of top fifty worldwide, and top five among the United States teams in the KiTS19 Competition.

Spatial Pyramid Pooling With 3D Convolution Improves Lung Cancer Detection.

Lung cancer is the leading cause of cancer deaths. Low-dose computed tomography (CT)screening has been shown to significantly reduce lung cancer mortality but suffers from a high false positive rate that leads to unnecessary diagnostic procedures. The development of deep learning techniques has the potential to help improve lung cancer screening technology. Here we present the algorithm, DeepScreener, which can predict a patient's cancer status from a volumetric lung CT scan. DeepScreener is based on our model of Spatial Pyramid Pooling, which ranked 16th of 1972 teams (top 1 percent)in the Data Science Bowl 2017 competition (DSB2017), evaluated with the challenge datasets. Here we test the algorithm with an independent set of 1449 low-dose CT scans of the National Lung Screening Trial (NLST)cohort, and we find that DeepScreener has consistent performance of high accuracy. Furthermore, by combining Spatial Pyramid Pooling and 3D Convolution, it achieves an AUC of 0.892, surpassing the previous state-of-the-art algorithms using only 3D convolution. The advancement of deep learning algorithms can potentially help improve lung cancer detection with low-dose CT scans.

A Continuously Benchmarked and Crowdsourced Challenge for Rapid Development and Evaluation of Models to Predict COVID-19 Diagnosis and Hospitalization.

Importance: Machine learning could be used to predict the likelihood of diagnosis and severity of illness. Lack of COVID-19 patient data has hindered the data science community in developing models to aid in the response to the pandemic. Objectives: To describe the rapid development and evaluation of clinical algorithms to predict COVID-19 diagnosis and hospitalization using patient data by citizen scientists, provide an unbiased assessment of model performance, and benchmark model performance on subgroups. Design, Setting, and Participants: This diagnostic and prognostic study operated a continuous, crowdsourced challenge using a model-to-data approach to securely enable the use of regularly updated COVID-19 patient data from the University of Washington by participants from May 6 to December 23, 2020. A postchallenge analysis was conducted from December 24, 2020, to April 7, 2021, to assess the generalizability of models on the cumulative data set as well as subgroups stratified by age, sex, race, and time of COVID-19 test. By December 23, 2020, this challenge engaged 482 participants from 90 teams and 7 countries. Main Outcomes and Measures: Machine learning algorithms used patient data and output a score that represented the probability of patients receiving a positive COVID-19 test result or being hospitalized within 21 days after receiving a positive COVID-19 test result. Algorithms were evaluated using area under the receiver operating characteristic curve (AUROC) and area under the precision recall curve (AUPRC) scores. Ensemble models aggregating models from the top challenge teams were developed and evaluated. Results: In the analysis using the cumulative data set, the best performance for COVID-19 diagnosis prediction was an AUROC of 0.776 (95% CI, 0.775-0.777) and an AUPRC of 0.297, and for hospitalization prediction, an AUROC of 0.796 (95% CI, 0.794-0.798) and an AUPRC of 0.188. Analysis on top models submitting to the challenge showed consistently better model performance on the female group than the male group. Among all age groups, the best performance was obtained for the 25- to 49-year age group, and the worst performance was obtained for the group aged 17 years or younger. Conclusions and Relevance: In this diagnostic and prognostic study, models submitted by citizen scientists achieved high performance for the prediction of COVID-19 testing and hospitalization outcomes. Evaluation of challenge models on demographic subgroups and prospective data revealed performance discrepancies, providing insights into the potential bias and limitations in the models.

Transfer learning with chest X-rays for ER patient classification.

One of the challenges with urgent evaluation of patients with acute respiratory distress syndrome (ARDS) in the emergency room (ER) is distinguishing between cardiac vs infectious etiologies for their pulmonary findings. We conducted a retrospective study with the collected data of 171 ER patients. ER patient classification for cardiac and infection causes was evaluated with clinical data and chest X-ray image data. We show that a deep-learning model trained with an external image data set can be used to extract image features and improve the classification accuracy of a data set that does not contain enough image data to train a deep-learning model. An analysis of clinical feature importance was performed to identify the most important clinical features for ER patient classification. The current model is publicly available with an interface at the web link: http://nbttranslationalresearch.org/ .

Highly accurate model for prediction of lung nodule malignancy with CT scans.

Computed tomography (CT) examinations are commonly used to predict lung nodule malignancy in patients, which are shown to improve noninvasive early diagnosis of lung cancer. It remains challenging for computational approaches to achieve performance comparable to experienced radiologists. Here we present NoduleX, a systematic approach to predict lung nodule malignancy from CT data, based on deep learning convolutional neural networks (CNN). For training and validation, we analyze >1000 lung nodules in images from the LIDC/IDRI cohort. All nodules were identified and classified by four experienced thoracic radiologists who participated in the LIDC project. NoduleX achieves high accuracy for nodule malignancy classification, with an AUC of ~0.99. This is commensurate with the analysis of the dataset by experienced radiologists. Our approach, NoduleX, provides an effective framework for highly accurate nodule malignancy prediction with the model trained on a large patient population. Our results are replicable with software available at http://bioinformatics.astate.edu/NoduleX .

DNAp: A Pipeline for DNA-seq Data Analysis.

Next-generation sequencing is empowering genetic disease research. However, it also brings significant challenges for efficient and effective sequencing data analysis. We built a pipeline, called DNAp, for analyzing whole exome sequencing (WES) and whole genome sequencing (WGS) data, to detect mutations from disease samples. The pipeline is containerized, convenient to use and can run under any system, since it is a fully automatic process in Docker container form. It is also open, and can be easily customized with user intervention points, such as for updating reference files and different software or versions. The pipeline has been tested with both human and mouse sequencing datasets, and it has generated mutations results, comparable to published results from these datasets, and reproducible across heterogeneous hardware platforms. The pipeline DNAp, funded by the US Food and Drug Administration (FDA), was developed for analyzing DNA sequencing data of FDA. Here we make DNAp an open source, with the software and documentation available to the public at http://bioinformatics.astate.edu/dna-pipeline/ .

SparRec: An effective matrix completion framework of missing data imputation for GWAS.

Genome-wide association studies present computational challenges for missing data imputation, while the advances of genotype technologies are generating datasets of large sample sizes with sample sets genotyped on multiple SNP chips. We present a new framework SparRec (Sparse Recovery) for imputation, with the following properties: (1) The optimization models of SparRec, based on low-rank and low number of co-clusters of matrices, are different from current statistics methods. While our low-rank matrix completion (LRMC) model is similar to Mendel-Impute, our matrix co-clustering factorization (MCCF) model is completely new. (2) SparRec, as other matrix completion methods, is flexible to be applied to missing data imputation for large meta-analysis with different cohorts genotyped on different sets of SNPs, even when there is no reference panel. This kind of meta-analysis is very challenging for current statistics based methods. (3) SparRec has consistent performance and achieves high recovery accuracy even when the missing data rate is as high as 90%. Compared with Mendel-Impute, our low-rank based method achieves similar accuracy and efficiency, while the co-clustering based method has advantages in running time. The testing results show that SparRec has significant advantages and competitive performance over other state-of-the-art existing statistics methods including Beagle and fastPhase.