Article
Critical Care Medicine
Phung Tran Huy Nhat, Nguyen Van Hao, Phan Vinh Tho, Hamideh Kerdegari, Luigi Pisani, Le Ngoc Minh Thu, Le Thanh Phuong, Ha Thi Hai Duong, Duong Bich Thuy, Angela McBride, Miguel Xochicale, Marcus Schultz, Reza Razavi, Andrew King, Louise Thwaites, Nguyen Van Vinh Chau, Sophie Yacoub, Alberto Gomez
Summary: This study developed an AI solution to assist clinicians in interpreting lung ultrasound images and evaluated its usefulness in a low resource ICU. The results showed that non-expert clinicians significantly improved their accuracy, time, and confidence in interpreting lung ultrasound images when using the AI tool.
Article
Computer Science, Artificial Intelligence
Nora El-Rashidy, Tamer Abuhmed, Louai Alarabi, Hazem M. El-Bakry, Samir Abdelrazek, Farman Ali, Shaker El-Sappagh
Summary: Sepsis is a life-threatening disease with difficulties in early identification, but establishing an accurate predictive model is crucial for improving prognosis.
NEURAL COMPUTING & APPLICATIONS
(2022)
Article
Critical Care Medicine
Gloria Hyunjung Kwak, Lowell Ling, Pan Hui
Summary: This study developed a model using deep learning to predict the need for vasopressors for critically ill patients within the first 24 hours of ICU admission based solely on vital signs. Results showed that respiratory rate, mean arterial pressure, and heart rate were the most important contributors to the model's predictions.
Article
Computer Science, Information Systems
Subhrajit Roy, Diana Mincu, Eric Loreaux, Anne Mottram, Ivan Protsyuk, Natalie Harris, Yuan Xue, Jessica Schrouff, Hugh Montgomery, Alistair Connell, Nenad Tomasev, Alan Karthikesalingam, Martin Seneviratne
Summary: The SeqSNR architecture demonstrated a modest yet statistically significant performance boost across 4 of the 6 tasks compared to single-task and naive multitasking approaches. When reducing the size of the training dataset for specific tasks, SeqSNR outperformed single-task in all cases, indicating superior label efficiency especially in scenarios where endpoint labels are difficult to ascertain.
JOURNAL OF THE AMERICAN MEDICAL INFORMATICS ASSOCIATION
(2021)
Article
Computer Science, Information Systems
Tingyi Wanyan, Akhil Vaid, Jessica K. De Freitas, Sulaiman Somani, Riccardo Miotto, Girish N. Nadkarni, Ariful Azad, Ying Ding, Benjamin S. Glicksberg
Summary: Traditional machine learning models have had limited success in predicting COVID-19 outcomes using EHR data, but a novel framework based on relational learning and heterogeneous graph model shows improved prediction accuracy. By leveraging diverse patient populations' EHR data in NYC, the model effectively captures patterns in patient representations of outcomes through relational learning strategy, leading to significant improvements in recall rates.
IEEE TRANSACTIONS ON BIG DATA
(2021)
Article
Health Care Sciences & Services
Steven Kessler, Dennis Schroeder, Sergej Korlakov, Vincent Hettlich, Sebastian Kalkhoff, Sobhan Moazemi, Artur Lichtenberg, Falko Schmid, Hug Aubin
Summary: In this article, a long short-term memory-based deep learning model (LSTM) is presented to assist physicians in deciding whether patients can be safely discharged from cardiovascular ICUs. The results show that the LSTM outperforms other models in terms of area under the receiver operating characteristic curve and precision-recall curve. This deep learning solution can help improve patient care and optimize ICU resources.
Article
Health Care Sciences & Services
Zhixuan Zeng, Yang Liu, Shuo Yao, Jiqiang Liu, Bing Xiao, Chenxue Liu, Xun Gong
Summary: This study proposes an attention architecture that is capable of predicting and handling missing data in electronic health records, showing strong model robustness.
Article
Computer Science, Interdisciplinary Applications
Ilaria Gandin, Arjuna Scagnetto, Simona Romani, Giulia Barbati
Summary: This study introduces an attention mechanism in LSTM neural network for predicting cardiovascular diseases, with a good performance measured by AUC of 0.790. The investigation on attention weights in the model's interpretability revealed that attention mechanisms can enhance the transparency of deep learning models in healthcare applications.
JOURNAL OF BIOMEDICAL INFORMATICS
(2021)
Article
Medical Informatics
Jiang Hu, Xiao-hui Kang, Fang-fang Xu, Ke-zhi Huang, Bin Du, Li Weng
Summary: This study applies machine learning approaches to predict life-threatening events in ICU patients using heterogeneous clinical data. The Light Gradient Boosting Machine showed the best performance, and short-term windows were more accurately predicted than medium-term windows. Features such as infusion pump related fluid input were highly related to life-threatening events. Furthermore, the prediction power of static features such as age and cardio-pulmonary function increased with the extended prediction window.
BMC MEDICAL INFORMATICS AND DECISION MAKING
(2022)
Article
Computer Science, Artificial Intelligence
Chunyang Cheng, Tianyang Xu, Xiao-Jun Wu
Summary: Existing image fusion approaches using a single network often yield suboptimal results due to the lack of ground-truth output. We propose a self-evolutionary training formula with a novel memory unit architecture (MUFusion) that utilizes intermediate fusion results for collaborative supervision. An adaptive unified loss function based on the memory unit is designed to improve fusion quality. Our MUFusion achieves superior performance in various image fusion tasks according to qualitative and quantitative experiments.
INFORMATION FUSION
(2023)
Article
Radiology, Nuclear Medicine & Medical Imaging
Jianhong Cheng, John Sollee, Celina Hsieh, Hailin Yue, Nicholas Vandal, Justin Shanahan, Ji Whae Choi, Thi My Linh Tran, Kasey Halsey, Franklin Iheanacho, James Warren, Abdullah Ahmed, Carsten Eickhoff, Michael Feldman, Eduardo Mortani Barbosa, Ihab Kamel, Cheng Ting Lin, Thomas Yi, Terrance Healey, Paul Zhang, Jing Wu, Michael Atalay, Harrison X. Bai, Zhicheng Jiao, Jianxin Wang
Summary: Deep learning models using longitudinal CXRs and clinical data were developed to predict in-hospital mortality for COVID-19 patients in the ICU. Models based on pre-ICU CXRs achieved an AUC of 0.632 and an accuracy of 0.593, models based on ICU CXRs achieved an AUC of 0.697 and an accuracy of 0.657, models based on all longitudinal CXRs achieved an AUC of 0.702 and an accuracy of 0.694, and models based on clinical data alone achieved an AUC of 0.653 and an accuracy of 0.657. The addition of longitudinal imaging to clinical data significantly improved mortality prediction, reaching an AUC of 0.727 and an accuracy of 0.732.
EUROPEAN RADIOLOGY
(2022)
Article
Critical Care Medicine
Chengfu Guan, Fuxin Ma, Sijie Chang, Jinhua Zhang
Summary: This study aimed to construct an interpretable machine learning model for predicting venous thromboembolism (VTE) in critically ill patients based on clinical features and laboratory indicators. The study found that the random forest model performed the best among all the constructed models.
Article
Pediatrics
Oi-Wa Chan, Wan-Hsuan Chen, Jainn-Jim Lin, Ming-Chou Chiang, Shao-Hsuan Hsia, Huei-Shyong Wang, En-Pei Lee, Yi-Shan Wang, Cheng-Yen Kuo, Kuang-Lin Lin
Summary: A clinical diagnosis by direct observation alone is insufficient for the identification of neonatal seizures. Continuous video-EEG monitoring plays a crucial role in diagnosing neonatal seizures and guiding clinical management decisions.
Article
Hematology
Li Luo, Ran Kou, Yuquan Feng, Jie Xiang, Wei Zhu
Summary: The study proposed a new diagnostic approach using machine learning prediction models to improve the intervention and diagnostic process of DVT in NICU patients. Two models with the strongest ROC were selected through cross-validation and ROC analysis, demonstrating the effectiveness of this predictive method.
CLINICAL AND APPLIED THROMBOSIS-HEMOSTASIS
(2021)
Article
Environmental Sciences
Praveen Kumar Sekharamantry, Farid Melgani, Jonni Malacarne
Summary: In this work, a deep learning-based scheme using Yolov5 architecture is proposed to detect apples in live apple farm images. The Yolov5 architecture is improved by incorporating an adaptive pooling scheme and attribute augmentation model, which detects smaller objects and improves the feature quality in complex backgrounds. A loss function is also incorporated to obtain accurate bounding boxes and maximize detection accuracy. The proposed approach achieves an overall accuracy of 0.97, 0.99, and 0.98 in terms of precision, recall, and F1-score, respectively.
Article
Clinical Neurology
E. Lotan, B. Zhang, S. Dogra, W. D. Wang, D. Carbone, G. Fatterpekar, E. K. Oermann, Y. W. Lui
Summary: This study introduced a deep learning model for automated preoperative and postoperative glioma segmentation, with a developed end-to-end pipeline for rapid and accurate processing. Results showed the model's excellent performance in whole-tumor segmentation, aiding clinicians in diagnosis and treatment.
AMERICAN JOURNAL OF NEURORADIOLOGY
(2022)
Article
Clinical Neurology
Nicholas L. Pitaro, Justin E. Tang, Varun Arvind, Brian H. Cho, Eric A. Geng, Uchechukwu O. Amakiri, Samuel K. Cho, Jun S. Kim
Summary: This retrospective cohort study compared readmission rates in surgically and non-surgically managed spinal epidural abscess (SEA) cases. The results showed that patients with a low comorbidity burden who underwent surgical management had significantly lower readmission rates than those managed non-surgically.
GLOBAL SPINE JOURNAL
(2023)
Review
Clinical Neurology
Justin E. Tang, Varun Arvind, Calista Dominy, Christopher A. White, Samuel K. Cho, Jun S. Kim
Summary: This study analyzes online written reviews of spine surgeons and explores the biases associated with demographic factors and trends in words utilized. The results show that gender and age have a significant impact on sentiment analysis scores. The best and worst reviewed surgeons are mainly evaluated based on behavioral factors and pain. The use of certain clinically relevant words affects the odds of a positive review. Therefore, establishing proper pain expectations prior to any intervention is crucial.
GLOBAL SPINE JOURNAL
(2023)
Review
Clinical Neurology
Akshar Patel, Christopher A. White, John T. Schwartz, Nicholas L. Pitaro, Kush C. Shah, Sirjanhar Singh, Varun Arvind, Jun S. Kim, Samuel K. Cho
Summary: New technologies such as machine learning, robot-guided spinal surgery, and patient-specific rods are increasingly being utilized to improve preoperative planning and patient satisfaction in the treatment of adult spinal deformity. These tools offer benefits such as predicting complications, improving screw placement accuracy, and reducing rod breakage rates, ultimately leading to better treatment outcomes.
Review
Clinical Neurology
Calista L. Dominy, Varun Arvind, Justin E. Tang, Christopher P. Bellaire, Sara Diana Pasik, Jun S. Kim, Samuel K. Cho
Summary: This study analyzed posts related to scoliosis surgery on social media platforms including Twitter, Instagram, and Reddit. It found that tweets were mostly positive, focusing on patient outcomes, while Instagram posts centered around post-operative progress and education resources. Reddit posts varied, with positive ones discussing personal progress and negative ones expressing fears and concerns about the surgery.
Article
Clinical Neurology
Justin E. Tang, Varun Arvind, Christopher A. White, Calista Dominy, Jun S. Kim, Samuel K. Cho
Summary: Physician review websites have a significant impact on patients' provider selection, and sentiment analysis through artificial intelligence can quantify surgeon reviews. A quantitative analysis of Scoliosis Research Society (SRS) members' written reviews showed a positive correlation between sentiment scores and star-rated reviews. Younger surgeons tended to receive more positive reviews, with pain management being a key factor in positive and negative reviews.
Article
Clinical Neurology
Aly A. Valliani, Nora C. Kim, Michael L. Martini, Jonathan S. Gal, Sean N. Neifert, Rui Feng, Eric A. Geng, Jun S. Kim, Samuel K. Cho, Eric K. Oermann, John M. Caridi
Summary: This study developed a robust machine learning algorithm to predict non-home discharge after thoracolumbar spine surgery, which demonstrated strong predictive ability across single-center and national patient cohorts. Important predictors identified by the algorithm included age, comorbidities, insurance type, and sex.
WORLD NEUROSURGERY
(2022)
Article
Orthopedics
Teja Yeramosu, Calista L. Dominy, Varun Arvind, Ula N. Isleem, Samuel K. Cho
Summary: The study analyzed posts on Twitter and Instagram related to scoliosis surgery to examine their tone, content, and perspectives. Instagram posts showed a positive tone and displayed patients' progress updates and contentment with surgery. Twitter posts, on the other hand, exhibited a negative tone, indicating discontentment towards inadequate access to surgery. The study suggests that surgeons can use social media platforms to connect with patients and provide information and awareness.
JOURNAL OF THE AMERICAN ACADEMY OF ORTHOPAEDIC SURGEONS
(2023)
Article
Multidisciplinary Sciences
Aly A. Valliani, Faris F. Gulamali, Young Joon Kwon, Michael L. Martini, Chiatse Wang, Douglas Kondziolka, Viola J. Chen, Weichung Wang, Anthony B. Costa, Eric K. Oermann
Summary: The fundamental challenge in machine learning is to ensure that trained models can effectively adapt to new and unseen data. Researchers have developed a general technique using generative adversarial networks (GANs) to mitigate the impact of dataset shift. The application of adversarial domain adaptation has led to improved model performance in tasks involving digit recognition and lung pathology classification.
Article
Orthopedics
Christopher A. White, Addison Quinones, Justin E. Tang, Liam R. Butler, Akiro H. Duey, Jun S. Kim, Samuel K. Cho, Paul J. Cagle
Summary: This study aimed to explore the impact of alcohol use disorder on readmissions and complications following total shoulder arthroplasty. It found that patients with an alcohol use disorder were more likely to experience shoulder dislocation, liver complications, and readmission within 90 days. Surgeons should therefore take caution and prevent complications and readmissions in patients with an alcohol use disorder during total shoulder arthroplasty.
JOURNAL OF ORTHOPAEDICS
(2023)
Article
Clinical Neurology
Gabrielle Price, Michael L. Martini, John M. Caridi, Darryl Lau, Eric K. Oermann, Sean N. Neifert
Summary: This study aimed to investigate the outcomes and risk profiles of multilevel fusion surgery for patients with Neurofibromatosis Type 1 (NF1). The study found that NF1 patients had certain complications following surgery, but there were no significant differences in quality or cost outcomes.
WORLD NEUROSURGERY
(2023)
Article
Orthopedics
Eric A. Geng, Brian H. Cho, Aly A. Valliani, Varun Arvind, Akshar V. Patel, Samuel K. Cho, Jun S. Kim, Paul J. Cagle
Summary: Demand for total shoulder arthroplasty is increasing, and the authors developed a machine learning model to identify shoulder implant manufacturers and types. Using convolutional neural network, the model achieved an accuracy of 93.9% on 696 X-ray images, assisting with preoperative planning and improving cost-efficiency in shoulder surgery.
JOURNAL OF ORTHOPAEDICS
(2023)
Editorial Material
Clinical Neurology
Samuel K. Cho