Internal Dataset Research Articles

Artificial intelligence-enhanced comprehensive assessment of the aortic valve stenosis continuum in echocardiography.

Transthoracic echocardiography (TTE) is the primary modality for diagnosing aortic stenosis (AS), yet it requires skilled operators and can be resource-intensive. We developed and validated an artificial intelligence (AI)-based system for evaluating AS that is effective in both resource-limited and advanced settings. We created a dual-pathway AI system for AS evaluation using a nationwide echocardiographic dataset (developmental dataset, n=8427): 1) a deep learning (DL)-based AS continuum assessment algorithm using limited 2D TTE videos, and 2) automating conventional AS evaluation. We performed internal (internal test dataset [ITDS], n=841) and external validation (distinct hospital dataset [DHDS], n=1696; temporally distinct dataset [TDDS], n=772) for diagnostic value across various stages of AS and prognostic value for composite endpoints (cardiovascular death, heart failure, and aortic valve replacement). The DL index for the AS continuum (DLi-ASc, range 0-100) increased with worsening AS severity and demonstrated excellent discrimination for any AS (AUC 0.91-0.99), significant AS (0.95-0.98), and severe AS (0.97-0.99). DLi-ASc was independent predictor for composite endpoint (adjusted hazard ratios 2.19, 1.64, and 1.61 per 10-point increase in ITDS, DHDS, and TDDS, respectively). Automatic measurement of conventional AS parameters demonstrated excellent correlation with manual measurement, resulting in high accuracy for ASstaging (98.2% for ITDS, 82.1% for DHDS, and 96.8% for TDDS) and comparable prognostic value to manually-derived parameters. The AI-based system provides accurate and prognostically valuable AS assessment, suitable for various clinical settings. Further validation studies are planned to confirm its effectiveness across diverse environments. This work was supported by a grant from the Institute of Information & Communications Technology Planning & Evaluation (IITP) funded by the Korea government (Ministry of Science and ICT; MSIT, Republic of Korea) (No. 2022000972, Development of a Flexible Mobile Healthcare Software Platform Using 5G MEC); and the Medical AI Clinic Program through the National IT Industry Promotion Agency (NIPA) funded by the MSIT, Republic of Korea (Grant No.: H0904-24-1002).

A Machine Learning-Based Radiomics Model for the Differential Diagnosis of Benign and Malignant Thyroid Nodules in F-18 FDG PET/CT: External Validation in the Different Scanner.

Accurate diagnosis is essential to avoid unnecessary procedures for thyroid incidentalomas (TIs). Advances in radiomics and machine learning applied to medical imaging offer promise for assessing thyroid nodules. This study utilized radiomics analysis on F-18 FDG PET/CT to improve preoperative differential diagnosis of TIs. A total of 152 patient cases were retrospectively analyzed and split into training and validation sets (7:3) using stratification and randomization. The least absolute shrinkage and selection operator (LASSO) algorithm identified nine radiomics features from 960 candidates to construct a radiomics signature predictive of malignancy. Performance of the radiomics score was evaluated using receiver operating characteristic (ROC) analysis and area under the curve (AUC). In the training set, the radiomics score achieved an AUC of 0.794 (95% CI: 0.703-0.885, p < 0.001). Validation was performed on internal and external datasets, yielding AUCs of 0.702 (95% CI: 0.547-0.858, p = 0.011) and 0.668 (95% CI: 0.500-0.838, p = 0.043), respectively. These results demonstrate that the selected nine radiomics features effectively differentiate malignant thyroid nodules. Overall, the radiomics model shows potential as a valuable predictive tool for thyroid cancer in patients with TIs, supporting improved preoperative decision-making.

Prediction of tissue and clinical thrombectomy outcome in acute ischaemic stroke using deep learning.

The advent of endovascular thrombectomy has significantly improved outcomes for stroke patients with intracranial large vessel occlusion, yet individual benefits can vary widely. As demand for thrombectomy rises and geographic disparities in stroke care access persist, there is a growing need for predictive models that quantify individual benefits. However, current imaging methods for estimating outcomes may not fully capture the dynamic nature of cerebral ischemia and lack a patient-specific assessment of thrombectomy benefits. Our study introduces a deep learning approach to predict individual responses to thrombectomy in acute ischemic stroke patients. The proposed models provide predictions for both tissue and clinical outcomes under two scenarios: one assuming successful reperfusion and another assuming unsuccessful reperfusion. The resulting simulations of penumbral salvage and difference in NIHSS at discharge quantify the potential individual benefits of the intervention. Our models were developed on an extensive dataset from routine stroke care, which included 405 ischemic stroke patients who underwent thrombectomy. We used acute data for training (n = 304), including multimodal CT imaging and clinical characteristics, along with post hoc markers like thrombectomy success, final infarct localization, and NIHSS at discharge. We benchmarked our tissue outcome predictions under the observed reperfusion scenario against a thresholding-based clinical method and a generalised linear model. Our deep-learning model showed significant superiority, with a mean Dice score of 0.48 on internal (n = 50) and 0.52 on external (n = 51) test data, versus 0.26/0.36 and 0.34/0.35 for the baselines, respectively. The NIHSS sum score prediction achieved median absolute errors of 1.5 NIHSS points on the internal test dataset and 3.0 NIHSS points on the external test dataset, outperforming other machine learning models. By predicting the patient-specific response to thrombectomy for both tissue and clinical outcomes, our approach offers an innovative biomarker that captures the dynamics of cerebral ischemia. We believe this method holds significant potential to enhance personalised therapeutic strategies and to facilitate efficient resource allocation in acute stroke care.

An open source convolutional neural network to detect and localize distal radius fractures on plain radiographs

PurposeDistal radius fractures (DRFs) are often initially assessed by junior doctors under time constraints, with limited supervision, risking significant consequences if missed. Convolutional Neural Networks (CNNs) can aid in diagnosing fractures. This study aims to internally and externally validate an open source algorithm for the detection and localization of DRFs.MethodsPatients from a level 1 trauma center from Adelaide, Australia that presented between 2016 and 2020 with wrist trauma were retrospectively included. Radiographs were reviewed confirming the presence or absence of a fracture, as well as annotating radius, ulna, and fracture location. An internal validation dataset from the same hospital was created. An external validation set was created with two other level 1 trauma centers, from Groningen and Rotterdam, the Netherlands. Three surgeons reviewed both sets for DRFs.ResultsThe algorithm was trained on 659 radiographs. The internal validation set included 190 patients, showing an accuracy of 87% and an AUC of 0.93 for DRF detection. The external validation set consisted of 188 patients, with an accuracy and AUC were 82% and 0.88 respectively. Radial and ulnar bone segmentation on the internal validation was excellent with an AP50 of 99 and 98, but moderate for fracture segmentation with an AP50 of 29. For external validation the AP50 was 92, 89 and 25 for radius, ulna, and fracture respectively.ConclusionThis open-source algorithm effectively detects DRFs with high accuracy and localizes them with moderate accuracy. It can assist clinicians in diagnosing suspected DRFs and is the first radiograph-based CNN externally validated on patients from multiple hospitals.

Exploring prognosis and therapeutic strategies for HBV-HCC patients based on disulfidptosis-related genes

BackgroundHepatocellular carcinoma (HCC) accounts for over 80% of primary liver cancers and is the third leading cause of cancer-related deaths worldwide. Hepatitis B virus (HBV) infection is the primary etiological factor. Disulfidptosis is a newly discovered form of regulated cell death. This study aims to develop a novel HBV-HCC prognostic signature related to disulfidptosis and explore potential therapeutic approaches through risk stratification based on disulfidptosis.MethodsTranscriptomic data from HBV-HCC patients were analyzed to identify BHDRGs. A prognostic model was established and validated using machine learning, with internal datasets and external datasets for verification. We then performed immune cell infiltration analysis, tumor microenvironment (TME) analysis, and immunotherapy-related analysis based on the prognostic signature. Besides, RT-qPCR and immunohistochemistry were conducted.ResultsA prognostic model was constructed using five genes (DLAT, STC2, POF1B, S100A9, and CPS1). A corresponding prognostic nomogram was developed based on riskScores, age, stage. Stratification by median risk score revealed a significant correlation between the prognostic signature and TME, tumor immune cell infiltration, immunotherapy efficacy, and drug sensitivity. The results of the experiments indicate that DLAT expression is higher in tumor tissues compared to adjacent tissues. DLAT expression is higher in HBV-HCC tumor tissues compared to normal tissues.ConclusionThis study stratifies HBV-HCC patients into distinct subgroups based on BHDRGs, establishing a prognostic model with significant implications for prognosis assessment, TME remodeling, and personalized therapy in HBV-HCC patients.

High-risk habitat radiomics model based on ultrasound images for predicting lateral neck lymph node metastasis in differentiated thyroid cancer

BackgroundThis study aims to evaluate the predictive usefulness of a habitat radiomics model based on ultrasound images for anticipating lateral neck lymph node metastasis (LLNM) in differentiated thyroid cancer (DTC), and for pinpointing high-risk habitat regions and significant radiomics traits.MethodsA group of 214 patients diagnosed with differentiated thyroid carcinoma (DTC) between August 2021 and August 2023 were included, consisting of 107 patients with confirmed postoperative lateral lymph node metastasis (LLNM) and 107 patients without metastasis or lateral cervical lymph node involvement. An additional cohort of 43 patients was recruited to serve as an independent external testing group for this study. Patients were randomly divided into training and internal testing group at an 8:2 ratio. Region of interest (ROI) was manually outlined, and habitat analysis subregions were defined using the K-means method. The ideal number of subregions (n = 5) was determined using the Calinski-Harabasz score, leading to the creation of a habitat radiomics model with 5 subregions and the identification of the high-risk habitat model. Area under the curve (AUC) values were calculated for all models to assess their validity, and predictive model nomograms were created by integrating clinical features. The internal and external testing dataset is employed to assess the predictive performance and stability of the model.ResultsIn internal testing group, Habitat 3 was identified as the high-risk habitat model in the study, showing the best diagnostic efficacy among all models (AUC(CRM) vs. AUC(Habitat 3) vs. AUC(CRM + Habitat 3) = 0.84(95%CI:0.71–0.97) vs. 0.90(95%CI:0.80-1.00) vs. 0.79(95%CI:0.65–0.93)). Moreover, integrating the Habitat 3 model with clinical features and constructing nomograms enhanced the predictive capability of the combined model (AUC = 0.95(95%CI:0.88-1.00)). In this study, an independent external testing cohort was utilized to assess the model’s accuracy, yielding an AUC of 0.88 (95%CI: 0.78–0.98).ConclusionThe integration of the High-Risk Habitats (Habitat 3) radiomics model with clinical characteristics demonstrated a high predictive accuracy in identifying LLNM. This model has the potential to offer valuable guidance to surgeons in deciding the necessity of LLNM dissection for DTC.Clinical trial numberNot applicable.

ADMET evaluation in drug discovery: 21. Application and industrial validation of machine learning algorithms for Caco-2 permeability prediction

The Caco-2 cell model has been widely used to assess the intestinal permeability of drug candidates in vitro, owing to its morphological and functional similarity to human enterocytes. While Caco-2 cell assay is considered safe and cost-effective, it is also characterized by being time-consuming. Therefore, computational models that achieve high accuracies in predicting Caco-2 permeability are crucial for enhancing the efficiency of oral drug development. In this study, we conducted an in-depth analysis of the characteristics of an augmented Caco-2 permeability dataset, and evaluated a diverse range of machine learning algorithms in combination with different molecular representations. The results indicated that XGBoost generally provided better predictions than comparable models for the test sets. In addition, we investigated the transferability of machine learning models trained on publicly available data to internal pharmaceutical industry datasets. Our findings, based on the Shanghai Qilu’s in-house dataset, showed that the boosting models retained a degree of predictive efficacy when applied to industry data. Furthermore, Y-randomization test and applicability domain analysis were employed to assess the robustness and generalizability of these models. Matched Molecular Pair Analysis (MMPA) was utilized to extract chemical transformation rules. We believe that the model developed in this study could represent a reliable tool for assessing Caco-2 permeability during early-stage drug discovery and the chemical transformation rules derived here could provide insights for optimizing Caco-2 permeability.Scientific contributionA comprehensive validation of various machine learning algorithms combined with diverse molecular representations on a large dataset for predicting Caco-2 permeability was reported. The transferability of machine learning models trained on publicly available data to internal pharmaceutical industry datasets was also investigated. Matched molecular pair analysis was carried out to provide reasonable suggestions for researchers to improve the Caco-2 permeability of compounds.Graphical

Deep learning model for identifying acute heart failure patients using electrocardiography in the emergency room.

Acute heart failure (AHF) poses significant diagnostic challenges in the emergency room (ER) because of its varied clinical presentation and limitations of traditional diagnostic methods. This study aimed to develop and evaluate a deep-learning model using electrocardiogram (ECG) data to enhance AHF identification in the ER. In this retrospective cohort study, we analyzed the ECG data of 19,285 patients who visited ERs of three hospitals between 2016 and 2020; 9,119 with available left ventricular ejection fraction and N-terminal prohormone of brain natriuretic peptide level data and who were diagnosed with AHF were included in the study. We extracted morphological and clinical parameters from ECG data to train and validate four machine learning models: baseline linear regression and more advanced models including XGBoost, Light GBM, and CatBoost. The CatBoost algorithm outperformed other models, showing superior area under the receiver operating characteristic and area under the precision-recall curve diagnostic accuracy across both internal (0.89 ± 0.01 and 0.89 ± 0.01) and external (0.90 and 0.89) validation datasets, respectively. The model demonstrated high accuracy, precision, recall, and f1 score, indicating robust performance in AHF identification. The developed machine learning model significantly enhanced AHF detection in the ER using conventional 12-lead ECGs combined with clinical data. These findings suggest that ECGs, a common tool in the ER, can effectively help screen for AHF.

Development and validation of a biomarker-based prediction model for metastasis in patients with colorectal cancer: Application of machine learning algorithms.

The purpose of the current study was to develop and validate a biomarker-based prediction model for metastasis in patients with colorectal cancer (CRC). Two datasets, GSE68468 and GSE41568, were retrieved from the Gene Expression Omnibus (GEO) database. In the GSE68468 dataset, key biomarkers were identified through a screening process involving differential expression analysis, redundancy analysis, and recursive feature elimination technique. Subsequently, the prediction model was developed and internally validated using five machine learning (ML) algorithms including lasso and elastic-net regularized generalized linear model (glmnet), k-nearest neighbors (kNN), support vector machine (SVM) with Radial Basis Function Kernel, random forest (RF), and eXtreme Gradient Boosting (XGBoost). The predictive performance of the algorithm with the highest accuracy was then externally validated on the GSE41568 dataset. Among 22,283 registered genes in the GSE68468 dataset, the screening process identified 16 key genes including MMP3, CCDC102B, CDH2, SCGB1A1, KRT7, CYP1B1, LAMC3, ALB, DIXDC1, VWF, MMP1, CYP4B1, NKX3-2, TMEM158, GADD45B, SERPINA1 and these genes were used to build the prediction model. On the internal validation dataset, the prediction performance of five ML algorithms was as follows; RF (accuracy=0.97 and kappa=0.91), XGBoost (0.93, 0.81), kNN (0.93, 0.81), glmnet (0.93, 0.82) and SVM (0.92, 0.80). Top five biomarkers were MMP3, CCDC102B, CDH2, VWF and MMP1. The RF model exhibited an accuracy of 0.97, a kappa value of 0.92, and an area under the curve (AUC) of 0.99 in the external validation dataset. The results of this study have identified biomarkers through ML algorithms which help to identify patients with CRC prone to metastasis.

Delta-Radiomics Using Machine Learning Classifiers With Auxiliary Data Sets to Predict Disease Progression During Magnetic Resonance-Guided Radiotherapy in Adrenal Metastases.

Adaptive radiotherapy accounts for interfractional anatomic changes. We hypothesize that changes in the gross tumor volumes identified during daily scans could be analyzed using delta-radiomics to predict disease progression events. We evaluated whether an auxiliary data set could improve prediction performance. We analyzed 108 patients (n = 90 internal; n = 18 external) who received ablative radiotherapy. The internal data set included 42 patients with adrenal cancer, 23 patients with lung cancer, and 25 patients with pancreatic cancer, with the clinical end point of progression-free survival events. The median dose was 50 Gy, which was delivered over five fractions. The delta features are the ratio of the features of the last to first treatment fraction, F5/F1, and the concatenation of the first and last fraction features, F1||F5. Decision tree classifier with and without auxiliary data sets, and the external data set was used exclusively for independent testing of the final models. During internal training, for the F1||F5 model, the inclusion of the lung data set increased our AUC receiver operator characteristic curve (ROC) from 0.53 ± 0.12 to 0.61 ± 0.11, whereas the pancreatic data set increased our AUC-ROC to 0.60 ± 0.14. For the F5/F1 model, the inclusion of the lung auxiliary data increased our AUC-ROC from 0.52 ± 0.13 to 0.65 ± 0.11, whereas it modestly changed by 0.62 ± 0.13 with the pancreas. During external testing, for the F5/F1 model, we reported an AUC-ROC of 0.60 with the lung auxiliary data and 0.43 with the pancreatic data. Also, for the F5||F1 model, we reported an AUC-ROC of 0.70 with the lung auxiliary and 0.60 with the pancreatic data. Decision trees provided an explainable model on the external data set. The validation of our model on an external data set may be the first step to biologically adapted radiotherapy recognizing radiomics signals for potential recurrence.

Pathomics-based machine learning models for predicting pathological complete response and prognosis in locally advanced rectal cancer patients post-neoadjuvant chemoradiotherapy: insights from two independent institutional studies

BackgroundAccurate prediction of pathological complete response (pCR) and disease-free survival (DFS) in locally advanced rectal cancer (LARC) patients undergoing neoadjuvant chemoradiotherapy (NCRT) is essential for formulating effective treatment plans. This study aimed to construct and validate the machine learning (ML) models to predict pCR and DFS using pathomics.MethodA retrospective analysis was conducted on 294 patients who received NCRT from two independent institutions. Pathomics from pre-NCRT H&E stains were extracted, and five ML models were developed and validated across two centers using ROC, Kaplan-Meier, time-dependent ROC, and nomogram analyses.ResultAmong the five ML models, the Xgboost (XGB) model demonstrated superior performance in predicting pCR, achieving an AUC of 1.000 (p < 0.001) on the internal data-set and an AUC of 0.950 (p = 0.001) on the external data-set.The XGB model effectively differentiated between high-risk and low-risk prognosis patients across all five centers: internal dataset (DFS, p = 0.002; OS, p = 0.004) and external dataset (DFS, p = 0.074; OS, p = 0.224).Furthermore, the COX regression demonstrated that the tumor length (HR = 1.230, 95%CI: 1.050–1.440, p = 0.010), post-NCRT CEA (HR = 1.716, 95%CI: 1.031– 2.858, p = 0.038), and XGB model score (HR = 0.128, 95%CI: 0.026–0.636, p = 0.012) were independent predictors of DFS after NCRT in the internal data-set.Using COX regression, the nomogram model and time-dependent AUC analysis demonstrated strong predictive discrimination for DFS in LARC patients across two independent institutions.ConclusionThe ML model based on pathomics demonstrated effective prediction of pCR and prognosis in LARC patients. Further validation in larger cohorts is warranted to confirm the findings of this study.

Development and validation of a risk prediction model for multiple organ dysfunction syndrome secondary to severe heat stroke based on immediate assessment indicators on ICU admission.

Early prediction of multiple organ dysfunction syndrome (MODS) secondary to severe heat stroke (SHS) is crucial for improving patient outcomes. This study aims to develop and validate a risk prediction model for those patients based on immediate assessment indicators on ICU admission. Two hundred eighty-four cases with SHS in our hospital between July 2009 and April 2024 were retrospectively reviewed, and categorized into non-MODS and MODS groups. Logistic regression analyses were performed to identify risk factors for MODS, and then to construct a risk prediction model, which was visualized by a nomogram. The predictive performance of the model was evaluated using the area under the receiver operating characteristic curve (AUC), Hosmer-Lemeshow (HL) test, calibration curve, and decision curve analysis (DCA). Finally, the AUCs of the prediction model was compared with other scoring systems. Acute gastrointestinal injury (AGI), heart rate (HR) >100 bpm, a decreased Glasgow Coma Scale (GCS) score, and elevated total bilirubin (TBil) within the first 24 h of ICU admission are identified as independent risk factors for the development of MODS in SHS patients. The model demonstrated good discriminative ability, and the AUC was 0.910 (95% CI: 0.856-0.965). Applying the predictive model to the internal validation dataset demonstrated good discrimination with an AUC of 0.933 (95% CI: 0.880-0.985) and good fit and calibration. The DCA of this model showed a superior clinical net benefit. The risk prediction model based on AGI, HR, GCS, and TBil shows robust predictive performance and clinical utility, which could serve as a reference for assessing and screening the risk of MODS in SHS patients.

M&M: an RNA-seq based pan-cancer classifier for paediatric tumours.

With many rare tumour types, acquiring the correct diagnosis is a challenging but crucial process in paediatric oncology. Historically, this is done based on histology and morphology of the disease. However, advances in genome wide profiling techniques such as RNA sequencing now allow the development of molecular classification tools. Here, we present M&M, a pan-paediatric cancer ensemble-based machine learning algorithm tailored towards inclusion of rare tumour types. The RNA-seq based algorithm can classify 52 different tumour types (precision ∼99%, recall ∼80%), plus the underlying 96 tumour subtypes (precision ∼96%, recall ∼70%). For low-confidence classifications, a comparable precision is achieved when including the three highest-scoring labels. We then validated M&M on an internal dataset (precision 99%, recall 76%) and an external dataset from the KidsFirst initiative (precision 98%, recall 77%). Finally, we show that M&M has similar performance as existing disease or domain specific classification algorithms based on RNA sequencing or methylation data. M&M's pan-cancer setup allows for easy clinical implementation, requiring only one classifier for all incoming diagnostic samples, including samples from different tumour stages and treatment statuses. Simultaneously, its performance is comparable to existing tumour- and tissue-specific classifiers. The introduction of an extensive pan-cancer classifier in diagnostics has the potential to increase diagnostic accuracy for many paediatric cancer cases, thereby contributing towards optimal patient survival and quality of life. Financial support was provided by the Foundation Children Cancer Free (KiKa core funding) and Adessium Foundation.

Uncertainty-aware automatic TNM staging classification for [18F] Fluorodeoxyglucose PET-CT reports for lung cancer utilising transformer-based language models and multi-task learning

Background[18F] Fluorodeoxyglucose (FDG) PET-CT is a clinical imaging modality widely used in diagnosing and staging lung cancer. The clinical findings of PET-CT studies are contained within free text reports, which can currently only be categorised by experts manually reading them. Pre-trained transformer-based language models (PLMs) have shown success in extracting complex linguistic features from text. Accordingly, we developed a multi-task ‘TNMu’ classifier to classify the presence/absence of tumour, node, metastasis (‘TNM’) findings (as defined by The Eight Edition of TNM Staging for Lung Cancer). This is combined with an uncertainty classification task (‘u’) to account for studies with ambiguous TNM status.Methods2498 reports were annotated by a nuclear medicine physician and split into train, validation, and test datasets. For additional evaluation an external dataset (n = 461 reports) was created, and annotated by two nuclear medicine physicians with agreement reached on all examples. We trained and evaluated eleven publicly available PLMs to determine which is most effective for PET-CT reports, and compared multi-task, single task and traditional machine learning approaches.ResultsWe find that a multi-task approach with GatorTron as PLM achieves the best performance, with an overall accuracy (all four tasks correct) of 84% and a Hamming loss of 0.05 on the internal test dataset, and 79% and 0.07 on the external test dataset. Performance on the individual TNM tasks approached expert performance with macro average F1 scores of 0.91, 0.95 and 0.90 respectively on external data. For uncertainty an F1 of 0.77 is achieved.ConclusionsOur ‘TNMu’ classifier successfully extracts TNM staging information from internal and external PET-CT reports. We concluded that multi-task approaches result in the best performance, and better computational efficiency over single task PLM approaches. We believe these models can improve PET-CT services by assisting in auditing, creating research cohorts, and developing decision support systems. Our approach to handling uncertainty represents a novel first step but has room for further refinement.

New insights into the role of mitophagy related gene affecting the metastasis of osteosarcoma through scRNA-seq and CRISPR-Cas9 genome editing

BackgroundOsteosarcoma (OSA), the most common primary bone malignancy, poses significant challenges due to its aggressive nature and propensity for metastasis, especially in adolescents. Mitophagy analysis can help identify new therapeutic targets and combined treatment strategies.MethodsThis study integrates single-cell sequencing (scRNA-seq) data and bulk-seq to identify mitophagy-related genes (MRGs) associated with the progression of OSA metastasis and analyze their clinical significance. scRNA-seq data elucidates the relationship between mitophagy and OSA metastasis, employing “CellChat” R package to explore intercellular communications and report on hundreds of ligand-receptor interactions. Subsequently, the combination of bulk-seq and CRISPR-Cas9 gene editing identifies mitophagy-related biomarker associated with metastatic prognosis. Finally, validation of the relationship between mitophagy and OSA metastasis is achieved through cellular biology experiments and animal studies.ResultsThe distinct mitophagy activity of various mitochondria manifests in diverse spatial localization, cellular developmental trajectories, and intercellular interactions. OSA tissue exhibits notable heterogeneity in mitophagy within osteoblastic OSA cells. However, high mitophagy activity correlates consistently with high metastatic potential. Subsequently, we identified three critical genes associated with mitophagy in OSA, namely RPS27A, TOMM20 and UBB. According to the aforementioned queue of genes, we have constructed a mitophagy_score (MIP_score). We observed that it consistently predicts patient prognosis in both internal and external datasets, demonstrating strong robustness and stability. Furthermore, we have found that MIP_score can also guide chemotherapy, with varying sensitivities to chemotherapeutic agents based on different MIP_score. It is noteworthy that, through the integration of CRISPR-Cas9 genome-wide screening and validation via cellular and animal experiments, we have identified RPS27A as a potential novel biomarker for OSA.ConclusionsOur comprehensive analysis elucidated the profile of mitophagy throughout the OSA metastasis process, forming the basis for a mitophagy-related prognostic model that addresses clinical outcomes and drug sensitivity following OSA metastasis. Additionally, an online interactive platform was established to assist clinicians in decision-making (https://mip-score.shinyapps.io/labtan/). These findings lay the groundwork for developing targeted therapies aimed at improving the prognosis of OSA patients.

MRAnnotator: multi-anatomy and many-sequence MRI segmentation of 44 structures

Abstract Purpose To develop a deep learning model for multi-anatomy segmentation of diverse anatomic structures on MRI. Materials and Methods In this retrospective study, 44 structures were annotated using a model-assisted workflow with manual human finalization in 2 curated datasets: an internal dataset of 1518 MRI sequences (843 patients) from various clinical sites within a health system, and an external dataset of 397 MRI sequences (263 patients) from an independent imaging center for benchmarking. The internal dataset was used to train an nnU-Net model (MRAnnotator), while the external dataset evaluated MRAnnotator’s generalizability across significant image acquisition distribution shifts. MRAnnotator was further benchmarked against an nnU-Net model trained on the AMOS dataset and 2 current multi-anatomy MRI segmentation models, TotalSegmentator MRI (TSM) and MRSegmentator (MRS). Performance throughout was quantified using the Dice score. Results MRAnnotator achieved an overall average Dice score of 0.878 (95% CI: 0.873, 0.884) on the internal dataset test set and 0.875 (95% CI: 0.869, 0.880) on the external dataset benchmark, demonstrating strong generalization (P = .899). On the AMOS test set, MRAnnotator achieved comparable performance for relevant classes (0.889 [0.866, 0.909]) to an AMOS-trained nnU-Net (0.895 [0.871, 0.915]) (P = .361) and outperformed TSM (0.822 [0.800, 0.842], P &amp;lt; .001) and MRS (0.867 [0.844, 0.887], P &amp;lt; .001). TSM and MRS were also evaluated on the relevant classes from the internal and external datasets and were unable to achieve comparable performance to MRAnnotator. Conclusion MRAnnotator achieves robust and generalizable MRI segmentation across 44 anatomic structures. Future direction will incorporate additional anatomic structures into the datasets and model. Model weights are publicly available on GitHub. The external test set with annotations is available upon request.

MRI-based multiregional radiomics for desmoplastic reaction classification and prognosis stratification in stage II rectal cancer: A bicenter study.

To develop an MRI-based multiregional radiomics model for the noninvasive desmoplastic reaction (DR) classification and prognosis stratification in stage II rectal cancer (RC) patients. This study retrospectively involved 336 patients with RC from two centers, with 239 from Center 1 divided into training (n=191) and internal validation (n=48) datasets at an 8:2 ratio, and 97 from Center 2 serving as external validation dataset. Radiomics features were extracted, and a multiregional radiomics DR (M-RDR) signature was established using multi-level feature selection procedure. The cut-off value for M-RDR was determined using Youden's index. We further evaluated the predictive values of M-RDR on prognosis and adjuvant chemotherapy stratification. The primary outcome was 3-year disease-free survival (DFS), and cox model performance was assessed using AUCs and 95% confidence intervals. M-RDR demonstrated a high accuracy in DR classification with AUCs of 0.778 and 0.798 in the training and internal validation datasets. Multivariable analysis confirmed M-RDR as an independent prognostic factor after adjusting for clinicopathological factors.The combined model incorporating M-RDR and clinicopathological factors showed good performance in predicting 3-year DFS, with AUCs of 0.923, 0.908, and 0.891 in the training, internal validation and external validation datasets, respectively. Additionally, patients in the M-RDR-high group who received adjuvant chemotherapy had significantly better DFS compared with those who did not (P<0.05). The MRI-based multiregional radiomics model could effectively improve non-invasive DR classification, and was able to enhance postoperative risk stratification and treatment decision-making in stage II RC patients.

BUSClean: Open-source software for breast ultrasound image pre-processing and knowledge extraction for medical AI.

Development of artificial intelligence (AI) for medical imaging demands curation and cleaning of large-scale clinical datasets comprising hundreds of thousands of images. Some modalities, such as mammography, contain highly standardized imaging. In contrast, breast ultrasound imaging (BUS) can contain many irregularities not indicated by scan metadata, such as enhanced scan modes, sonographer annotations, or additional views. We present an open-source software solution for automatically processing clinical BUS datasets. The algorithm performs BUS scan filtering (flagging of invalid and non-B-mode scans), cleaning (dual-view scan detection, scan area cropping, and caliper detection), and knowledge extraction (BI-RADS Labeling and Measurement fields) from sonographer annotations. Its modular design enables users to adapt it to new settings. Experiments on an internal testing dataset of 430 clinical BUS images achieve >95% sensitivity and >98% specificity in detecting every type of text annotation, >98% sensitivity and specificity in detecting scans with blood flow highlighting, alternative scan modes, or invalid scans. A case study on a completely external, public dataset of BUS scans found that BUSClean identified text annotations and scans with blood flow highlighting with 88.6% and 90.9% sensitivity and 98.3% and 99.9% specificity, respectively. Adaptation of the lesion caliper detection method to account for a type of caliper specific to the case study demonstrates the intended use of BUSClean in new data distributions and improved performance in lesion caliper detection from 43.3% and 93.3% out-of-the-box to 92.1% and 92.3% sensitivity and specificity, respectively. Source code, example notebooks, and sample data are available at https://github.com/hawaii-ai/bus-cleaning.

Optimal thresholds and key parameters for predicting influenza A virus transmission events in ferrets

Although assessments of influenza A virus transmissibility in the ferret model play a critical role in pandemic risk evaluations, few studies have investigated which virological data collected from virus-inoculated animals are most predictive of subsequent virus transmission to naïve contacts. We compiled viral titer data from >475 ferrets inoculated with 97 contemporary IAV (including high- and low-pathogenicity avian, swine-origin, and human viruses of multiple HA subtypes) that served as donors for assessments of virus transmission in the presence of direct contact (DCT) or via respiratory droplets (RDT). A diversity of molecular determinants, clinical parameters, and infectious titer measurements and derived quantities were examined to identify which metrics were most statistically supported with transmission outcome. Higher viral loads in nasal wash (NW) specimens were strongly associated with higher transmission frequencies in DCT, but not RDT models. However, viruses that reached peak titers in NW specimens early (day 1 p.i.) were strongly associated with higher transmission in both models. Interestingly, viruses with ‘intermediate’ transmission outcomes (33–66%) had NW titers and derived quantities more similar to non-transmissible viruses (<33%) in a DCT setting, but with efficiently transmissible viruses (>67%) in a RDT setting. Machine learning was employed to further assess the predictive role of summary measures and varied interpretation of intermediate transmission outcomes in both DCT and RDT models, with models employing these different thresholds yielding high performance metrics against both internal and external datasets. Collectively, these findings suggest that higher viral load in inoculated animals can be predictive of DCT outcomes, whereas the timing of when peak titers are detected in inoculated animals can inform RDT outcomes. Identification that intermediate transmission outcomes should be contextualized relative to the transmission mode assessed provides needed refinement towards improving interpretation of ferret transmission studies in the context of pandemic risk assessment.

Self-supervised learning improves robustness of deep learning lung tumor segmentation models to CT imaging differences.

Self-supervised learning (SSL) is an approach to extract useful feature representations from unlabeled data, and enable fine-tuning on downstream tasks with limited labeled examples. Self-pretraining is a SSL approach that uses curated downstream task dataset for both pretraining and fine-tuning. Availability of large, diverse, and uncurated public medical image sets presents the opportunity to potentially create foundation models by applying SSL in the "wild" that are robust to imaging variations. However, the benefit of wild- versus self-pretraining has not been studied for medical imageanalysis. Compare robustness of wild versus self-pretrained models created using convolutional neural network (CNN) and transformer (vision transformer [ViT] and hierarchical shifted window [Swin]) models for non-small cell lung cancer (NSCLC) segmentation from 3D computed tomography (CT) scans. CNN, ViT, and Swin models were wild-pretrained using unlabeled 10,412 3D CTs sourced from the cancer imaging archive and internal datasets. Self-pretraining was applied to same networks using a curated public downstream task dataset (n = 377) of patients with NSCLC. Pretext tasks introduced in self-distilled masked image transformer were used for both pretraining approaches. All models were fine-tuned to segment NSCLC (n = 377 training dataset) and tested on two separate datasets containing early (public n = 156) and advanced stage (internal n = 196) NSCLC. Models were evaluated in terms of: (a) accuracy, (b) robustness to image differences from contrast, slice thickness, and reconstruction kernels, and (c) impact of pretext tasks for pretraining. Feature reuse was evaluated using centered kernel alignment. Wild-pretrained Swin models resulted in higher feature reuse at earlier level layers and increased feature differentiation close to output. Wild-pretrained Swin outperformed self-pretrained models for analyzed imaging acquisitions. Neither ViT nor CNN showed a clear benefit of wild-pretraining compared to self-pretraining. Masked image prediction pretext task that forces networks to learn the local structure resulted in higher accuracy compared to contrastive task that models global image information. Wild-pretrained Swin networks were more robust to analyzed CT imaging differences for lung tumor segmentation than self-pretrained methods. ViT and CNN models did not show a clear benefit for wild-pretraining overself-pretraining.