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

Deep learning (DL) algorithms could improve the classification of ovarian tumours assessed with ultrasound (US) images and clinical information. This study aimed to develop a DL model for diagnosing ovarian tumours based on US images and clinical information, and to compare the performance of the DL model with radiologist assessment. This study retrospectively collected data from four hospitals involving women who underwent US examinations for ovarian tumours. Additionally, data were prospectively randomly collected from two other hospitals. Histopathological analysis served as the reference standard. The retrospective dataset was divided into training, test and internal validation sets. The prospective dataset was used as an external validation set. A DL model was developed. The performance of the DL model was compared with the eleven radiologists’ assessments in the internal and external validation sets. The performance of radiologists' assessments with or without DL model were also compared. A total of 1518 women (6432 images) with ovarian tumours were collected, including 364 women in the internal validation dataset and 340 women in the external validation dataset. In the external validation dataset, the accuracy for the DL model's classification of benign, borderline, and malignant ovarian tumours were 85.5%, which were comparable to the expert assessments (85.3%, p > 0.05), but significantly higher than the average level of eleven different-experience radiologists (72.5%, p < 0.05). The diagnostic performance of junior and mid-level experienced radiologists were significantly improved after applying the DL model. The DL model developed using US images and clinical information can effectively classify benign, malignant, and borderline ovarian tumours with diagnostic performance comparable to the expert assessment and improving junior and mid-level experienced radiologists.

Background: The number of patients diagnosed with rectal neuroendocrine tumors (R-NETs) is increasing year by year. An integrated survival predictive model is required to predict the prognosis of R-NETs. The present study is aimed at exploring epidemiological characteristics of R-NETs based on a retrospective study from the Surveillance, Epidemiology, and End Results (SEER) database and predicting survival of R-NETs with machine learning.Methods: Data of patients with R-NETs were extracted from the SEER database (2000–2017), and data were also retrospectively collected from a single medical center in China. The main outcome measure was the 5-year survival status. Risk factors affecting survival were analyzed by Cox regression analysis, and six common machine learning algorithms were chosen to build the predictive models. Data from the SEER database were divided into a training set and an internal validation set according to the year 2010 as a time point. Data from China were chosen as an external validation set. The best machine learning predictive model was compared with the American Joint Committee on Cancer (AJCC) seventh staging system to evaluate its predictive performance in the internal validation dataset and external validation dataset.Results: A total of 10,580 patients from the SEER database and 68 patients from a single medical center were included in the analysis. Age, gender, race, histologic type, tumor size, tumor number, summary stage, and surgical treatment were risk factors affecting survival status. After the adjustment of parameters and algorithms comparison, the predictive model using the eXtreme Gradient Boosting (XGBoost) algorithm had the best predictive performance in the training set [area under the curve (AUC) = 0.87, 95%CI: 0.86–0.88]. In the internal validation, the predictive ability of XGBoost was better than that of the AJCC seventh staging system (AUC: 0.90 vs. 0.78). In the external validation, the XGBoost predictive model (AUC = 0.89) performed better than the AJCC seventh staging system (AUC = 0.83).Conclusions: The XGBoost algorithm had better predictive power than the AJCC seventh staging system, which had a potential value of the clinical application.

Small intracranial aneurysms are being increasingly detected while the rupture risk is not well-understood. We aimed to develop rupture-risk models of small aneurysms by combining clinical, morphologic, and hemodynamic information based on machine learning techniques and to test the models in external validation datasets. From January 2010 to December 2016, five hundred four consecutive patients with only small aneurysms (<5 mm) detected by CTA and invasive cerebral angiography (or surgery) were retrospectively enrolled and randomly split into training (81%) and internal validation (19%) sets to derive and validate the proposed machine learning models (support vector machine, random forest, logistic regression, and multilayer perceptron). Hemodynamic parameters were obtained using computational fluid dynamics simulation. External validation was performed in other hospitals to test the models. The support vector machine performed the best with areas under the curve of 0.88 (95% CI, 0.85-0.92) and 0.91 (95% CI, 0.74-0.98) in the training and internal validation datasets, respectively. Feature ranks suggested hemodynamic parameters, including stable flow pattern, concentrated inflow streams, and a small (<50%) flow-impingement zone, and the oscillatory shear index coefficient of variation, were the best predictors of aneurysm rupture. The support vector machine showed an area under the curve of 0.82 (95% CI, 0.69-0.94) in the external validation dataset, and no significant difference was found for the areas under the curve between internal and external validation datasets (P = .21). This study revealed that machine learning had a good performance in predicting the rupture status of small aneurysms in both internal and external datasets. Aneurysm hemodynamic parameters were regarded as the most important predictors.

Deep learning model for diagnosing gastric mucosal lesions using endoscopic images: development, validation, and method comparison

A deep learning algorithm for detection of oral cavity squamous cell carcinoma from photographic images: A retrospective study.

Illness severity assessment of older adults in critical illness using machine learning (ELDER-ICU): an international multicentre study with subgroup bias evaluation

BackgroundThere remains a demand for a practical method of identifying lipid-poor adrenal lesions.PurposeTo explore the predictive value of computed tomography (CT) features combined with demographic characteristics for lipid-poor adrenal adenomas and nonadenomas.Materials and MethodsWe retrospectively recruited patients with lipid-poor adrenal lesions between January 2015 and August 2021 from two independent institutions as follows: Institution 1 for the training set and the internal validation set and Institution 2 for the external validation set. Two radiologists reviewed CT images for the three sets. We performed a least absolute shrinkage and selection operator (LASSO) algorithm to select variables; subsequently, multivariate analysis was used to develop a generalized linear model. The probability threshold of the model was set to 0.5 in the external validation set. We calculated the sensitivity, specificity, accuracy, and area under the receiver operating characteristic curve (AUC) for the model and radiologists. The model was validated and tested in the internal validation and external validation sets; moreover, the accuracy between the model and both radiologists were compared using the McNemar test in the external validation set.ResultsIn total, 253 patients (median age, 55 years [interquartile range, 47–64 years]; 135 men) with 121 lipid-poor adrenal adenomas and 132 nonadenomas were included in Institution 1, whereas another 55 patients were included in Institution 2. The multivariable analysis showed that age, male, lesion size, necrosis, unenhanced attenuation, and portal venous phase attenuation were independently associated with adrenal adenomas. The clinical-image model showed AUCs of 0.96 (95% confidence interval [CI]: 0.91, 0.98), 0.93 (95% CI: 0.84, 0.97), and 0.86 (95% CI: 0.74, 0.94) in the training set, internal validation set, and external validation set, respectively. In the external validation set, the model showed a significantly and non-significantly higher accuracy than reader 1 (84% vs. 65%, P = 0.031) and reader 2 (84% vs. 69%, P = 0.057), respectively.ConclusionsOur clinical-image model displayed good utility in differentiating lipid-poor adrenal adenomas. Further, it showed better diagnostic ability than experienced radiologists in the external validation set.

A CT radiomics-based machine learning model for predicting high-grade pathological components in stage IA lung adenocarcinoma: a two-center study.

Computerized image analysis for whole-slide images has been shown to improve efficiency, accuracy, and consistency in histopathology evaluations. We aimed to assess whether immunohistochemistry (IHC) image quantitative features can reflect the immune status and provide prognostic information for colorectal cancer patients. A fully automated pipeline was designed to extract histogram features from IHC digital images in a training set (N = 243). A Hist-Immune signature was generated with selected features using the LASSO Cox model. The results were validated using internal (N = 147) and external (N = 76) validation sets. The five-feature-based Hist-Immune signature was significantly associated with overall survival in training (HR 2.72, 95% CI 1.68–4.41, P < .001), internal (2.86, 1.28–6.39, 0.010), and external (2.30, 1.02–6.16, 0.044) validation sets. The full model constructed by integrating the Hist-Immune signature and clinicopathological factors had good discrimination ability (C-index 0.727, 95% CI 0.678–0.776), confirmed using internal (0.703, 0.621–0.784) and external (0.756, 0.653–0.859) validation sets. Our findings indicate that the Hist-Immune signature constructed based on the quantitative features could reflect the immune status of patients with colorectal cancer, which might advocate change in risk stratification and consequent precision medicine.

BackgroundIn clinical practice, some patients with hyperthyroidism or hypothyroidism may lack typical clinical manifestations. Due to its popularity, this study sought to explore whether the computed tomography (CT) examination method could be used to conduct radiomics research to identify such patients. Specifically, this study aimed to investigate the value of radiomics based on non-contrast CT in the auxiliary diagnosis of hyperthyroidism/hypothyroidism.MethodsThe data of 1,181 patients at The Second Affiliated Hospital of Soochow University were retrospectively collected. The patients were randomly divided into a training group (n=827) and an internal validation group (n=354) at a ratio of 7:3. Additionally, the data of 415 patients from Changshu Hospital Affiliated to Nanjing University of Chinese Medicine were collected to serve as the external validation set. Radiomics features were extracted from non-contrast CT images, and the features related to hyperthyroidism/hypothyroidism were selected by a significance analysis and Pearson correlation analysis. A clinical model was then built based on clinical indicators, and a combined model was developed by integrating radiomics-predicted risk values with clinical indicators. The model’s three-class results were analyzed using a confusion matrix, and its binary classification diagnosis results were assessed using receiver operating characteristic (ROC) curves.ResultsA total of 21 radiomics features and four clinical features were screened for modeling. The average F1 scores and kappa values for the three-classification diagnosis of thyroid function in the clinical, radiomics, and combined models were 0.664, 0.709, 0.739 and 0.493, 0.556, and 0.609 in the internal validation set, and 0.540, 0.660, 0.697, and 0.310, 0.483, and 0.531 in the external validation set, respectively. The ROC curve analysis showed that the combined model outperformed the clinical and radiomics model in the two-classification diagnosis in both the internal and external validation sets.ConclusionsThe predictive model based on non-contrast CT radiomics had good diagnostic efficacy for hyperthyroidism/hypothyroidism. Its accuracy was further improved by combining clinical information. This model could alert clinicians to patients lacking typical clinical manifestations.

Accurate early diagnosis of ovarian cancer is crucial. The objective of this research is to create a comprehensive model that merges clinical variables, O-RADS, and deep learning radiomics to support preoperative diagnosis and assess its efficacy for sonographers. Data from two centers were used: Center 1 for training and internal validation, and Center 2 for external validation. DL and radiomics features were extracted from transvaginal ultrasound images to create a DL radiomics model using the LASSO method. A machine learning model ensemble was created by merging clinical variables, O-RADS scores, and DL radiomics model predictions. The model's effectiveness was evaluated by measuring the area under the receiver operating characteristic curve (AUC) and analyzing its impact on improving the diagnostic skills of sonographers. Moreover, the model's additional usefulness was assessed through integrated discrimination improvement (IDI), net reclassification improvement (NRI), and subgroup analysis. The ensemble model demonstrated superior diagnostic performance for ovarian cancer compared to standalone clinical models and clinical O-RADS models. Notably, there were significant improvements in the NRI and IDI across all three datasets, with p-values < 0.05. The ensemble model exhibited exceptional diagnostic performance, achieving AUCs of 0.97 in both the internal and external validation sets. Moreover, the implementation of this ensemble model substantially improved the diagnostic precision and reliability of sonographers. The sonographers' average AUC improved by 11 % in the internal validation set and by 7.7 % in the external validation set. The ensemble model significantly enhances preoperative ovarian cancer diagnosis accuracy and improves sonographers' diagnostic capabilities and consistency.

Background Incidence rate of thyroid cancer is steadily increasing due to overdiagnosis and overtreatment. Thyroid ultrasound is commonly used to diagnose thyroid cancer. The aim of this study is to examine the accuracy of using deep convolutional neural network (DCNN) models to improve diagnosis of thyroid cancer by analyzing sonographic imaging data from clinical thyroid ultrasound. Methods A total of 131,731 sonographic images from 17,627 thyroid cancer patients and 180,668 sonographic images from 25,325 controls used as training set were obtained from Tianjin Cancer Hospital. Images from anatomical sites that did not have cancer according to location sign on the image were not included. All thyroid cancer patients and 13·2% of controls (51,255 images) were confirmed by pathological reports. DCNN is a specific type of neural network optimized for image recognition. We trained two DCNN models on the training set and subsequently evaluated the performance on one independent internal (Tianjin, 1,118 individuals) and two external (Jilin,154 individuals; Weihai, 1,420 individuals) validation sets. Individuals in the validation sets all have pathological examinations. We compared the specificity/sensitivity of DCNN models with the performance of six thyroid ultrasound radiologists on these three validation sets. Findings DCNN model achieved high performance in identifying thyroid cancer patients versus six experience radiologists: for Tianjin validation set, sensitivity was 92·2% versus 96·9% (95% CI 89·7% - 94·3% vs. 93·9% - 98·6%; p = 0·003), and specificity was 85·6% versus 59·4% (95% CI 82·4% - 88·4% vs. 53% - 65·6%; p &amp;lt; 0·0001); for Jilin validation set, sensitivity was 84·3% versus 92·9% (95% CI 73·6% - 91·9% vs. 84·1% - 97·6%; p = 0·05), and specificity was 86·9% versus 57·1% (95% CI 77·8% - 93·3% vs. 45·9% - 67·9%; p &amp;lt; 0·0001); for Weihai validation set, sensitivity was 84·5% versus 89% (95% CI 81·2% - 87·4% vs. 81·9% - 94%; p = 0·2), and specificity was 87·5% versus 68·6% (95% CI 85·1% - 89·6% vs. 60·7% - 75·8%; p &amp;lt; 0·0001). Interpretation DCNN models exhibited high accuracy, sensitivity, and specificity in identifying thyroid cancer patients at levels comparable to or higher than six experienced radiologists. Conferred by the high specificity of DCNN models, the rate of overdiagnosis and overtreatment of patients with thyroid cancer is expected to decrease. This supports future application of the deep learning models to clinical practice for thyroid cancer diagnosis. However, further validation of these DCNN models in prospective clinical trials is warranted. Funding The Program for Changjiang Scholars and Innovative Research Team in University in China (IRT_14R40), National Natural Science Foundation of China (31801117). Citation Format: Xiangchun Li, Sheng Zhang, Qiang Zhang, Xi Wei, Yi Pan, Jing Zhao, Xiaojie Xin, Xiaoqing Wang, Fan Yang, Jianxin Li, Meng Yang, Qinghua Wang, Xiangqian Zheng, Yanhui Zhao, Lun Zhang, Xudong Wang, Zhimin Zheng, Christopher T. Whitlow, Metin N. Gurcan, Boris Pasche, Ming Gao, Wei Zhang, Kexin Chen. Diagnosis of thyroid cancer using deep convolutional neural network models applied to sonographic images from clinical ultrasound exams [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 1394.

To test the technical reproducibility of acquisition and scanners of CT image-based radiomics model for early recurrent hepatocellular carcinoma (HCC). We included primary HCC patient undergone curative therapies, using early recurrence as endpoint. Four datasets were constructed: 109 images from hospital #1 for training (set 1: 1-mm image slice thickness), 47 images from hospital #1 for internal validation (sets 2 and 3: 1-mm and 10-mm image slice thicknesses, respectively), and 47 images from hospital #2 for external validation (set 4: vastly different from training dataset). A radiomics model was constructed. Radiomics technical reproducibility was measured by overfitting and calibration deviation in external validation dataset. The influence of slice thickness on reproducibility was evaluated in two internal validation datasets. Compared with set 1, the model in set 2 indicated favorable prediction efficiency (the area under the curve 0.79 vs. 0.80, P = 0.47) and good calibration (unreliability statistic U: P = 0.33). However, in set 4, significant overfitting (0.63 vs. 0.80, P < 0.01) and calibration deviation (U: P < 0.01) were observed. Similar poor performance was also observed in set 3 (0.56 vs. 0.80, P = 0.02; U: P < 0.01). CT-based radiomics has poor reproducibility between centers. Image heterogeneity, such as slice thickness, can be a significant influencing factor.

IntroductionAsthma is one of the commonest chronic conditions in the world. Subtypes of asthma have been defined, typically from clinical datasets on small, well-characterised subpopulations of asthma patients. We sought to define asthma subtypes from large longitudinal primary care electronic health records (EHRs) using cluster analysis. MethodsIn this retrospective cohort study, we extracted asthma subpopulations from the Optimum Patient Care Research Database (OPCRD) to robustly train and test algorithms, and externally validated findings in the Secure Anonymised Information Linkage (SAIL) Databank. In both databases, we identified adults with an asthma diagnosis code recorded in the three years prior to an index date. Train and test datasets were selected from OPCRD using an index date of Jan 1, 2016. Two internal validation datasets were selected from OPCRD using index dates of Jan 1, 2017 and 2018. Three external validation datasets were selected from SAIL using index dates of Jan 1, 2016, 2017 and 2018. Each dataset comprised 50,000 randomly selected non-overlapping patients. Subtypes were defined by applying multiple correspondence analysis and k-means cluster analysis to the train dataset, and were validated in the internal and external validation datasets. ResultsWe defined six asthma subtypes with clear clinical interpretability: low inhaled corticosteroid (ICS) use and low healthcare utilisation (30% of patients); low-to-medium ICS use (36%); low-to-medium ICS use and comorbidities (12%); varied ICS use and comorbid chronic obstructive pulmonary disease (4%); high (10%) and very high ICS use (7%). The subtypes were replicated with high accuracy in internal (91–92%) and external (84–86%) datasets. ConclusionAsthma subtypes derived and validated in large independent EHR databases were primarily defined by level of ICS use, level of healthcare use, and presence of comorbidities. This has important clinical implications towards defining asthma subtypes, facilitating patient stratification, and developing more personalised monitoring and treatment strategies.

Background and aim Direct oral anticoagulants (DOACs) carry a risk of gastrointestinal bleeding (GIB). We aimed to develop and validate machine learning (ML) models to predict GIB in DOAC users and compare them with conventional risk scores. Methods We retrospectively analyzed 4,494 patients aged ≥18 years prescribed DOACs from December 2014 to October 2020. Patients were allocated to the training (n = 3,147), internal (n = 677), and external (n = 670) validation cohorts. Three ML algorithms, Gradient Boosting Machine (GBM), XGBoost, and Generalized Linear Model (GLM), predicted GIB at 12 and 24 months. Performance was assessed using the area under the receiver operating characteristic curve (AUC) and specificity at 100% sensitivity, compared with the HAS-BLED, ATRIA, VTE-BLEED, and ORBIT scores. Results At 24 months, XGBoost achieved the AUCs in the training (0.862), internal validation (0.819), and external validation (0.905) sets. At 12 months, XGBoost performed with AUCs of 0.917, 0.839, and 0.948, respectively. XGBoost exceeded the conventional scores, although ORBIT was the best among the latter (AUC 0.780 at 24 months, 0.728 at 12 months). The ML models also achieved higher specificity at 100% sensitivity. At 12 months, XGBoost and GB model demonstrated 79.8% specificity at 100% sensitivity, whereas GLM showed 67.8%. The conventional models were lower, with an ORBIT of 39.8%. By 24 months, GLM and ORBIT specificities were 43.8% and 40.0%, respectively. Conclusions ML models, particularly XGBoost, outperformed traditional bleeding risk scores in predicting GIB in DOAC users. However, the performance of the ML models was unsatisfactory. Further research is warranted to achieve a better performance.