Two-photon synthetic aperture microscopy for minimally invasive fast 3D imaging of native subcellular behaviors in deep tissue.

Holistic understanding of physio-pathological processes requires noninvasive 3D imaging in deep tissue across multiple spatial and temporal scales to link diverse transient subcellular behaviors with long-term physiogenesis. Despite broad applications of two-photon microscopy (TPM), there remains an inevitable tradeoff among spatiotemporal resolution, imaging volumes, and durations due to the point-scanning scheme, accumulated phototoxicity, and optical aberrations. Here, we harnessed the concept of synthetic aperture radar in TPM to achieve aberration-corrected 3D imaging of subcellular dynamics at a millisecond scale for over 100,000 large volumes in deep tissue, with three orders of magnitude reduction in photobleaching. With its advantages, we identified direct intercellular communications through migrasome generation following traumatic brain injury, visualized the formation process of germinal center in the mouse lymph node, and characterized heterogeneous cellular states in the mouse visual cortex, opening up a horizon for intravital imaging to understand the organizations and functions of biological systems at a holistic level.

Evaluating GPT-V4 (GPT-4 with Vision) on Detection of Radiologic Findings on Chest Radiographs.

Background Generating radiologic findings from chest radiographs is pivotal in medical image analysis. The emergence of OpenAI's generative pretrained transformer, GPT-4 with vision (GPT-4V), has opened new perspectives on the potential for automated image-text pair generation. However, the application of GPT-4V to real-world chest radiography is yet to be thoroughly examined. Purpose To investigate the capability of GPT-4V to generate radiologic findings from real-world chest radiographs. Materials and Methods In this retrospective study, 100 chest radiographs with free-text radiology reports were annotated by a cohort of radiologists, two attending physicians and three residents, to establish a reference standard. Of 100 chest radiographs, 50 were randomly selected from the National Institutes of Health (NIH) chest radiographic data set, and 50 were randomly selected from the Medical Imaging and Data Resource Center (MIDRC). The performance of GPT-4V at detecting imaging findings from each chest radiograph was assessed in the zero-shot setting (where it operates without prior examples) and few-shot setting (where it operates with two examples). Its outcomes were compared with the reference standard with regards to clinical conditions and their corresponding codes in the International Statistical Classification of Diseases, Tenth Revision (ICD-10), including the anatomic location (hereafter, laterality). Results In the zero-shot setting, in the task of detecting ICD-10 codes alone, GPT-4V attained an average positive predictive value (PPV) of 12.3%, average true-positive rate (TPR) of 5.8%, and average F1 score of 7.3% on the NIH data set, and an average PPV of 25.0%, average TPR of 16.8%, and average F1 score of 18.2% on the MIDRC data set. When both the ICD-10 codes and their corresponding laterality were considered, GPT-4V produced an average PPV of 7.8%, average TPR of 3.5%, and average F1 score of 4.5% on the NIH data set, and an average PPV of 10.9%, average TPR of 4.9%, and average F1 score of 6.4% on the MIDRC data set. With few-shot learning, GPT-4V showed improved performance on both data sets. When contrasting zero-shot and few-shot learning, there were improved average TPRs and F1 scores in the few-shot setting, but there was not a substantial increase in the average PPV. Conclusion Although GPT-4V has shown promise in understanding natural images, it had limited effectiveness in interpreting real-world chest radiographs. © RSNA, 2024 Supplemental material is available for this article.

A span-based model for extracting overlapping PICO entities from randomized controlled trial publications.

OBJECTIVES: Extracting PICO (Populations, Interventions, Comparison, and Outcomes) entities is fundamental to evidence retrieval. We present a novel method, PICOX, to extract overlapping PICO entities. MATERIALS AND METHODS: PICOX first identifies entities by assessing whether a word marks the beginning or conclusion of an entity. Then, it uses a multi-label classifier to assign one or more PICO labels to a span candidate. PICOX was evaluated using 1 of the best-performing baselines, EBM-NLP, and 3 more datasets, ie, PICO-Corpus and randomized controlled trial publications on Alzheimer's Disease (AD) or COVID-19, using entity-level precision, recall, and F1 scores. RESULTS: PICOX achieved superior precision, recall, and F1 scores across the board, with the micro F1 score improving from 45.05 to 50.87 (P ≪.01). On the PICO-Corpus, PICOX obtained higher recall and F1 scores than the baseline and improved the micro recall score from 56.66 to 67.33. On the COVID-19 dataset, PICOX also outperformed the baseline and improved the micro F1 score from 77.10 to 80.32. On the AD dataset, PICOX demonstrated comparable F1 scores with higher precision when compared to the baseline. CONCLUSION: PICOX excels in identifying overlapping entities and consistently surpasses a leading baseline across multiple datasets. Ablation studies reveal that its data augmentation strategy effectively minimizes false positives and improves precision.

A natural language processing approach to detect inconsistencies in death investigation notes attributing suicide circumstances.

BACKGROUND: Data accuracy is essential for scientific research and policy development. The National Violent Death Reporting System (NVDRS) data is widely used for discovering the patterns and causing factors of death. Recent studies suggested the annotation inconsistencies within the NVDRS and the potential impact on erroneous suicide-circumstance attributions. METHODS: We present an empirical Natural Language Processing (NLP) approach to detect annotation inconsistencies and adopt a cross-validation-like paradigm to identify possible label errors. We analyzed 267,804 suicide death incidents between 2003 and 2020 from the NVDRS. We measured annotation inconsistency by the degree of changes in the F-1 score. RESULTS: Our results show that incorporating the target state's data into training the suicide-circumstance classifier brings an increase of 5.4% to the F-1 score on the target state's test set and a decrease of 1.1% on other states' test set. CONCLUSIONS: To conclude, we present an NLP framework to detect the annotation inconsistencies, show the effectiveness of identifying and rectifying possible label errors, and eventually propose an improvement solution to improve the coding consistency of human annotators.

Data accuracy is essential for scientific research and policy development. The National Violent Death Reporting System (NVDRS) contains the recording of individual suicide incidents taking place in the United States, and the contributing suicide circumstances. We used a computational method to check the accuracy of NVDRS records. Our method identified and rectified possible errors in labeling within the database. This method could be used to improve the label accuracy in the NVDRS database, enabling more accurate recording and study of suicide circumstances. Improved data recording of suicide circumstances could potentially be used to develop improved approaches to prevent suicide in the future.

Closing the gap between open-source and commercial large language models for medical evidence summarization.

Large language models (LLMs) hold great promise in summarizing medical evidence. Most recent studies focus on the application of proprietary LLMs. Using proprietary LLMs introduces multiple risk factors, including a lack of transparency and vendor dependency. While open-source LLMs allow better transparency and customization, their performance falls short compared to proprietary ones. In this study, we investigated to what extent fine-tuning open-source LLMs can further improve their performance in summarizing medical evidence. Utilizing a benchmark dataset, MedReview, consisting of 8,161 pairs of systematic reviews and summaries, we fine-tuned three broadly-used, open-sourced LLMs, namely PRIMERA, LongT5, and Llama-2. Overall, the fine-tuned LLMs obtained an increase of 9.89 in ROUGE-L (95% confidence interval: 8.94-10.81), 13.21 in METEOR score (95% confidence interval: 12.05-14.37), and 15.82 in CHRF score (95% confidence interval: 13.89-16.44). The performance of fine-tuned LongT5 is close to GPT-3.5 with zero-shot settings. Furthermore, smaller fine-tuned models sometimes even demonstrated superior performance compared to larger zero-shot models. The above trends of improvement were also manifested in both human and GPT4-simulated evaluations. Our results can be applied to guide model selection for tasks demanding particular domain knowledge, such as medical evidence summarization.

Closing the gap between open source and commercial large language models for medical evidence summarization.

Large language models (LLMs) hold great promise in summarizing medical evidence. Most recent studies focus on the application of proprietary LLMs. Using proprietary LLMs introduces multiple risk factors, including a lack of transparency and vendor dependency. While open-source LLMs allow better transparency and customization, their performance falls short compared to the proprietary ones. In this study, we investigated to what extent fine-tuning open-source LLMs can further improve their performance. Utilizing a benchmark dataset, MedReview, consisting of 8161 pairs of systematic reviews and summaries, we fine-tuned three broadly-used, open-sourced LLMs, namely PRIMERA, LongT5, and Llama-2. Overall, the performance of open-source models was all improved after fine-tuning. The performance of fine-tuned LongT5 is close to GPT-3.5 with zero-shot settings. Furthermore, smaller fine-tuned models sometimes even demonstrated superior performance compared to larger zero-shot models. The above trends of improvement were manifested in both a human evaluation and a larger-scale GPT4-simulated evaluation.

Hidden Flaws Behind Expert-Level Accuracy of Multimodal GPT-4 Vision in Medicine.

Recent studies indicate that Generative Pre-trained Transformer 4 with Vision (GPT-4V) outperforms human physicians in medical challenge tasks. However, these evaluations primarily focused on the accuracy of multi-choice questions alone. Our study extends the current scope by conducting a comprehensive analysis of GPT-4V's rationales of image comprehension, recall of medical knowledge, and step-by-step multimodal reasoning when solving New England Journal of Medicine (NEJM) Image Challenges - an imaging quiz designed to test the knowledge and diagnostic capabilities of medical professionals. Evaluation results confirmed that GPT-4V performs comparatively to human physicians regarding multi-choice accuracy (81.6% vs. 77.8%). GPT-4V also performs well in cases where physicians incorrectly answer, with over 78% accuracy. However, we discovered that GPT-4V frequently presents flawed rationales in cases where it makes the correct final choices (35.5%), most prominent in image comprehension (27.2%). Regardless of GPT-4V's high accuracy in multi-choice questions, our findings emphasize the necessity for further in-depth evaluations of its rationales before integrating such multimodal AI models into clinical workflows.

Hidden flaws behind expert-level accuracy of multimodal GPT-4 vision in medicine.

Recent studies indicate that Generative Pre-trained Transformer 4 with Vision (GPT-4V) outperforms human physicians in medical challenge tasks. However, these evaluations primarily focused on the accuracy of multi-choice questions alone. Our study extends the current scope by conducting a comprehensive analysis of GPT-4V's rationales of image comprehension, recall of medical knowledge, and step-by-step multimodal reasoning when solving New England Journal of Medicine (NEJM) Image Challenges-an imaging quiz designed to test the knowledge and diagnostic capabilities of medical professionals. Evaluation results confirmed that GPT-4V performs comparatively to human physicians regarding multi-choice accuracy (81.6% vs. 77.8%). GPT-4V also performs well in cases where physicians incorrectly answer, with over 78% accuracy. However, we discovered that GPT-4V frequently presents flawed rationales in cases where it makes the correct final choices (35.5%), most prominent in image comprehension (27.2%). Regardless of GPT-4V's high accuracy in multi-choice questions, our findings emphasize the necessity for further in-depth evaluations of its rationales before integrating such multimodal AI models into clinical workflows.

Towards long-tailed, multi-label disease classification from chest X-ray: Overview of the CXR-LT challenge.

Many real-world image recognition problems, such as diagnostic medical imaging exams, are "long-tailed" - there are a few common findings followed by many more relatively rare conditions. In chest radiography, diagnosis is both a long-tailed and multi-label problem, as patients often present with multiple findings simultaneously. While researchers have begun to study the problem of long-tailed learning in medical image recognition, few have studied the interaction of label imbalance and label co-occurrence posed by long-tailed, multi-label disease classification. To engage with the research community on this emerging topic, we conducted an open challenge, CXR-LT, on long-tailed, multi-label thorax disease classification from chest X-rays (CXRs). We publicly release a large-scale benchmark dataset of over 350,000 CXRs, each labeled with at least one of 26 clinical findings following a long-tailed distribution. We synthesize common themes of top-performing solutions, providing practical recommendations for long-tailed, multi-label medical image classification. Finally, we use these insights to propose a path forward involving vision-language foundation models for few- and zero-shot disease classification.

Towards long-tailed, multi-label disease classification from chest X-ray: Overview of the CXR-LT challenge.

Many real-world image recognition problems, such as diagnostic medical imaging exams, are "long-tailed" - there are a few common findings followed by many more relatively rare conditions. In chest radiography, diagnosis is both a long-tailed and multi-label problem, as patients often present with multiple findings simultaneously. While researchers have begun to study the problem of long-tailed learning in medical image recognition, few have studied the interaction of label imbalance and label co-occurrence posed by long-tailed, multi-label disease classification. To engage with the research community on this emerging topic, we conducted an open challenge, CXR-LT, on long-tailed, multi-label thorax disease classification from chest X-rays (CXRs). We publicly release a large-scale benchmark dataset of over 350,000 CXRs, each labeled with at least one of 26 clinical findings following a long-tailed distribution. We synthesize common themes of top-performing solutions, providing practical recommendations for long-tailed, multi-label medical image classification. Finally, we use these insights to propose a path forward involving vision-language foundation models for few- and zero-shot disease classification.

Real-time denoising enables high-sensitivity fluorescence time-lapse imaging beyond the shot-noise limit.

A fundamental challenge in fluorescence microscopy is the photon shot noise arising from the inevitable stochasticity of photon detection. Noise increases measurement uncertainty and limits imaging resolution, speed and sensitivity. To achieve high-sensitivity fluorescence imaging beyond the shot-noise limit, we present DeepCAD-RT, a self-supervised deep learning method for real-time noise suppression. Based on our previous framework DeepCAD, we reduced the number of network parameters by 94%, memory consumption by 27-fold and processing time by a factor of 20, allowing real-time processing on a two-photon microscope. A high imaging signal-to-noise ratio can be acquired with tenfold fewer photons than in standard imaging approaches. We demonstrate the utility of DeepCAD-RT in a series of photon-limited experiments, including in vivo calcium imaging of mice, zebrafish larva and fruit flies, recording of three-dimensional (3D) migration of neutrophils after acute brain injury and imaging of 3D dynamics of cortical ATP release. DeepCAD-RT will facilitate the morphological and functional interrogation of biological dynamics with a minimal photon budget.

Printed microfluidic sweat sensing platform for cortisol and glucose detection.

Wearable sweat biosensors offer compelling opportunities for improved personal health monitoring and non-invasive measurements of key biomarkers. Inexpensive device fabrication methods are necessary for scalable manufacturing of portable, disposable, and flexible sweat sensors. Furthermore, real-time sweat assessment must be analyzed to validate measurement reliability at various sweating rates. Here, we demonstrate a "smart bandage" microfluidic platform for cortisol detection and continuous glucose monitoring integrated with a synthetic skin. The low-cost, laser-cut microfluidic device is composed of an adhesive-based microchannel and solution-processed electrochemical sensors fabricated from inkjet-printed graphene and silver solutions. An antibody-derived cortisol sensor achieved a limit of detection of 10 pM and included a low-voltage electrowetting valve, validating the microfluidic sensor design under typical physiological conditions. To understand effects of perspiration rate on sensor performance, a synthetic skin was developed using soft lithography to mimic human sweat pores and sweating rates. The enzymatic glucose sensor exhibited a range of 0.2 to 1.0 mM, a limit of detection of 10 µM, and reproducible response curves at flow rates of 2.0 µL min-1 and higher when integrated with the synthetic skin, validating its relevance for human health monitoring. These results demonstrate the potential of using printed microfluidic sweat sensors as a low-cost, real-time, multi-diagnostic device for human health monitoring.

Three-Dimensional CeO2 Woodpile Nanostructures To Enhance Performance of Enzymatic Glucose Biosensors.

Fabrication of detection elements with ultrahigh surface area is essential for improving the sensitivity of analyte detection. Here, we report a direct patterning technique to fabricate three-dimensional CeO2 nanoelectrode arrays for biosensor application over relatively large areas. The fabrication approach, which employs nanoimprint lithography and a CeO2 nanoparticle-based ink, enables the direct, high-throughput patterning of nanostructures and is scalable, integrable, and of low cost. With the convenience of sequential imprinting, multilayered woodpile nanostructures with prescribed numbers of layers were achieved in a "stacked-up" architecture and were successfully fabricated over large areas. To demonstrate application as a biosensor, an enzymatic glucose sensor was developed. The sensitivity of glucose sensors can be enhanced simply by increasing the number of layers, which multiplies surface area while maintaining a constant footprint. The four-layer woodpile nanostructure of CeO2 glucose sensor exhibited enhanced sensitivity (42.8 µA mM-1 cm-2) and good selectivity. This direct imprinting strategy for three-dimensional sensing architectures is potentially extendable to other electroactive materials and other sensing applications.

Wavelength-Selective Three-Dimensional Thermal Emitters via Imprint Lithography and Conformal Metallization.

Metallic photonic crystals (MPCs) exhibit wavelength-selective thermal emission enhancements and are promising thermal optical devices for various applications. Here, we report a scalable fabrication strategy for MPCs suitable for high-temperature applications. Well-defined double-layer titanium dioxide (TiO2) woodpile structures are fabricated using a layer-by-layer soft-imprint method with TiO2 nanoparticle ink dispersions, and the structures are subsequently coated with high purity, conformal gold films via reactive deposition from supercritical carbon dioxide. The resulting gold-coated woodpile structures are effective MPCs and exhibit emissivity enhancements at a selective wavelength. Gold coatings deposited using a cold-wall reactor are found to be smoother and result in a greater thermal emission enhancement compared to those deposited using a hot-wall reactor.

Experimental verification of a two-dimensional respiratory motion compensation system with ultrasound tracking technique in radiation therapy.

This study proposed respiratory motion compensation system (RMCS) combined with an ultrasound image tracking algorithm (UITA) to compensate for respiration-induced tumor motion during radiotherapy, and to address the problem of inaccurate radiation dose delivery caused by respiratory movement. This study used an ultrasound imaging system to monitor respiratory movements combined with the proposed UITA and RMCS for tracking and compensation of the respiratory motion. Respiratory motion compensation was performed using prerecorded human respiratory motion signals and also sinusoidal signals. A linear accelerator was used to deliver radiation doses to GAFchromic EBT3 dosimetry film, and the conformity index (CI), root-mean-square error, compensation rate (CR), and planning target volume (PTV) were used to evaluate the tracking and compensation performance of the proposed system. Human respiratory pattern signals were captured using the UITA and compensated by the RMCS, which yielded CR values of 34-78%. In addition, the maximum coronal area of the PTV ranged from 85.53 mm2 to 351.11 mm2 (uncompensated), which reduced to from 17.72 mm2 to 66.17 mm2 after compensation, with an area reduction ratio of up to 90%. In real-time monitoring of the respiration compensation state, the CI values for 85% and 90% isodose areas increased to 0.7 and 0.68, respectively. The proposed UITA and RMCS can reduce the movement of the tracked target relative to the LINAC in radiation therapy, thereby reducing the required size of the PTV margin and increasing the effect of the radiation dose received by the treatment target.

Direct Imprinting of Scalable, High-Performance Woodpile Electrodes for Three-Dimensional Lithium-Ion Nanobatteries.

The trend of device downscaling drives a corresponding need for power source miniaturization. Though numerous microfabrication methods lead to successful creation of submillimeter-scale electrodes, scalable approaches that provide cost-effective nanoscale resolution for energy storage devices such as on-chip batteries remain elusive. Here, we report nanoimprint lithography (NIL) as a direct patterning technique to fabricate high-performance TiO2 nanoelectrode arrays for lithium-ion batteries (LIBs) over relatively large areas. The critical electrode dimension is below 200 nm, which enables the structure to possess favorable rate capability even under discharging current densities as high as 5000 mA g-1. In addition, by sequential imprinting, electrodes with three-dimensional (3D) woodpile architecture were readily made in a "stack-up" manner. The height of architecture can be easily controlled by the number of stacked layers while maintaining nearly constant surface-to-volume ratios. The result is a proportional increase of areal capacity with the number of layers. The structure-processing combination leads to efficient use of the material, and the resultant specific capacity (250.9 mAh g-1) is among the highest reported. This work provides a simple yet effective strategy to fabricate nanobatteries and can be potentially extended to other electroactive materials.

UV-nanoimprint lithography as a tool to develop flexible microfluidic devices for electrochemical detection.

Research in microfluidic biosensors has led to dramatic improvements in sensitivities. Very few examples of these devices have been commercially successful, keeping this methodology out of the hands of potential users. In this study, we developed a method to fabricate a flexible microfluidic device containing electrowetting valves and electrochemical transduction. The device was designed to be amenable to a roll-to-roll manufacturing system, allowing a low manufacturing cost. Microchannels with high fidelity were structured on a PET film using UV-NanoImprint Lithography (UV-NIL). The electrodes were inkjet-printed and photonically sintered on second flexible PET film. The film containing electrodes was bonded directly to the channel-containing layer to form sealed fluidic device. Actuation of the multivalve system with food dye in PBS buffer was performed to demonstrate automated fluid delivery. The device was then used to detect Salmonella in a liquid sample.