Benchmark Tests of Atom Segmentation Deep Learning Models with a Consistent Dataset.

The information content of atomic-resolution scanning transmission electron microscopy (STEM) images can often be reduced to a handful of parameters describing each atomic column, chief among which is the column position. Neural networks (NNs) are high performance, computationally efficient methods to automatically locate atomic columns in images, which has led to a profusion of NN models and associated training datasets. We have developed a benchmark dataset of simulated and experimental STEM images and used it to evaluate the performance of two sets of recent NN models for atom location in STEM images. Both models exhibit high performance for images of varying quality from several different crystal lattices. However, there are important differences in performance as a function of image quality, and both models perform poorly for images outside the training data, such as interfaces with large difference in background intensity. Both the benchmark dataset and the models are available using the Foundry service for dissemination, discovery, and reuse of machine learning models.

Graph network based deep learning of bandgaps.

Recent machine learning models for bandgap prediction that explicitly encode the structure information to the model feature set significantly improve the model accuracy compared to both traditional machine learning and non-graph-based deep learning methods. The ongoing rapid growth of open-access bandgap databases can benefit such model construction not only by expanding their domain of applicability but also by requiring constant updating of the model. Here, we build a new state-of-the-art multi-fidelity graph network model for bandgap prediction of crystalline compounds from a large bandgap database of experimental and density functional theory (DFT) computed bandgaps with over 806 600 entries (1500 experimental, 775 700 low-fidelity DFT, and 29 400 high-fidelity DFT). The model predicts bandgaps with a 0.23 eV mean absolute error in cross validation for high-fidelity data, and including the mixed data from all different fidelities improves the prediction of the high-fidelity data. The prediction error is smaller for high-symmetry crystals than for low symmetry crystals. Our data are published through a new cloud-based computing environment, called the "Foundry," which supports easy creation and revision of standardized data structures and will enable cloud accessible containerized models, allowing for continuous model development and data accumulation in the future.

14 examples of how LLMs can transform materials science and chemistry: a reflection on a large language model hackathon.

Large-language models (LLMs) such as GPT-4 caught the interest of many scientists. Recent studies suggested that these models could be useful in chemistry and materials science. To explore these possibilities, we organized a hackathon. This article chronicles the projects built as part of this hackathon. Participants employed LLMs for various applications, including predicting properties of molecules and materials, designing novel interfaces for tools, extracting knowledge from unstructured data, and developing new educational applications. The diverse topics and the fact that working prototypes could be generated in less than two days highlight that LLMs will profoundly impact the future of our fields. The rich collection of ideas and projects also indicates that the applications of LLMs are not limited to materials science and chemistry but offer potential benefits to a wide range of scientific disciplines.

FAIR principles for AI models with a practical application for accelerated high energy diffraction microscopy.

A concise and measurable set of FAIR (Findable, Accessible, Interoperable and Reusable) principles for scientific data is transforming the state-of-practice for data management and stewardship, supporting and enabling discovery and innovation. Learning from this initiative, and acknowledging the impact of artificial intelligence (AI) in the practice of science and engineering, we introduce a set of practical, concise, and measurable FAIR principles for AI models. We showcase how to create and share FAIR data and AI models within a unified computational framework combining the following elements: the Advanced Photon Source at Argonne National Laboratory, the Materials Data Facility, the Data and Learning Hub for Science, and funcX, and the Argonne Leadership Computing Facility (ALCF), in particular the ThetaGPU supercomputer and the SambaNova DataScale® system at the ALCF AI Testbed. We describe how this domain-agnostic computational framework may be harnessed to enable autonomous AI-driven discovery.

Inspissated cyst fluid in endoscopic ultrasound-guided fine needle aspiration of pancreatic cysts.

BACKGROUND: Inspissated cyst fluid may be identified on pancreatic cyst aspiration cytology. We report on the cytomorphologic characteristics of inspissated cyst fluid on EUS-FNA of pancreatic cysts and correlate this finding with histopathology or multimodal (cytology, cyst fluid analysis, molecular pathology, imaging) classification of cyst type. METHODS: The department archives were searched for pancreatic cyst fine-needle aspiration biopsies that contained dessicated, crystalline or inspissated material on cytologic preparations. RESULTS: Twenty-eight cases of pancreatic cysts containing inspissated material were identified. The cytomorphology of the inspissated material ranged from fibrillary fan-like structure (54%), ball-like structures (57%), and granular material (43%). When present, the fibrillary inspissated material was associated with neoplastic mucinous cysts in 11/15 (73%) cases, but was also seen in 2 pseudocysts, 1 serous cystadenoma, and 1 cyst of uncertain type (suggestive of mucinous cyst on EUS). The presence of fibrillary inspissated cyst material on cytology had a positive predictive value of 79% and a specificity of 63% for a neoplastic mucinous cyst. CONCLUSION: Although not highly specific, the presence of inspissated cyst fluid with fibrillary architecture should be recognized by cytopathologists and interpreted as at least "atypical" given the potential association with neoplastic mucinous cysts of the pancreas.

The histologic evolution of revealed, acute abruptions.

There is considerable interest in using pathology to confirm acute abruptions. It has been suggested that pathologic findings can help to determine the timing of abruptions. Because of the dearth of evidence in the literature supporting this claim and its medicolegal implications, we undertook this study to explore further the possibility of timing abruptions by histopathology. We sought to correlate bleeding interval (duration from maternal presentation with vaginal bleeding [revealed abruption] to placental delivery) with placental histopathologic findings. We performed a retrospective review of clinical data and placental pathology from all cases of clinically diagnosed, acute, revealed abruptions at a single, large institution in New England between 2000 and 2015. Cases were identified based on clinical diagnoses, bleeding intervals were calculated from clinical notes, and histologic evaluations were performed by 2 pathologists blinded to the bleeding intervals. A total of 177 cases were analyzed. Of these, 103 (58%) had histologic findings corroborating the clinical diagnosis of abruption. The most frequent finding was maternal surface indentation (51 cases) followed by intravillous hemorrhage (50 cases). The former was also the earliest finding, with a minimum bleeding interval of 4 minutes. In multivariate modeling, plasma cell deciduitis was significantly associated with a longer bleeding interval (median 63 hours). If there were 2 pathologic findings, there was a trend toward a longer bleeding interval. There was modest sensitivity for the pathologic diagnosis of acute revealed abruption. Although there was not a clear, stepwise progression of histologic lesions; the presence of 2 or more findings tended to be seen with longer bleeding intervals. Our results suggest that histologic findings cannot be used to time acute revealed abruptions reliably, and any interpretation of such should be made with caution.

Preoperative characteristics and cytological features of 136 histologically confirmed pancreatic mucinous cystic neoplasms.

BACKGROUND: Mucinous cystic neoplasms (MCNs) of the pancreas present a management conundrum. The majority are benign but all are resected due to their malignant potential. Recent studies have recommended nonsurgical management. In the current study, the authors analyzed the preoperative imaging, cytology, and cyst fluid characteristics of 136 histologically confirmed MCNs to assess predictors of a high-risk (HR) cyst for surgical triage. METHODS: MCNs resected at the Massachusetts General Hospital between 1990 and 2014 formed the study cohort. Patient demographics, cyst size, and mural nodules (MNs) by endoscopic ultrasound, cytology, and cyst fluid carcinoembryonic antigen and amylase levels were correlated with histological grade. A HR cyst was defined as high-grade dysplasia or invasive carcinoma on histology. Performance characteristics were assessed for each parameter, with a cyst size ≥3 cm or a MN on imaging and malignant cytology considered to be "true-positive" results for predicting malignancy. RESULTS: Only 15 of the 136 cysts had HR histology (11%). On average, patients with HR cysts were older than those with low-risk cysts (55 years vs 49 years, respectively). High-grade cytology was the most accurate predictor of malignancy (95%) followed by MN and cyst size together (88%) and MN alone (83%). The average carcinoembryonic antigen level (in ng/mL) increased with the grade of dysplasia but the ranges overlapped between low risk and HR cysts. CONCLUSIONS: To the authors' knowledge, the current study is the largest series to date analyzing the cytological features of histologically confirmed MCN. Cytology is insensitive but very specific for detecting a HR MCN and outperformed imaging for the detection of HR MCN. Endoscopic ultrasound-guided fine-needle aspiration and cytology should be performed on any clinically suspected MCN that is being considered for conservative management. Cancer Cytopathol 2017;125:169-177. © 2016 American Cancer Society.