Construction and clinical validation of a fetal brain magnetic resonance imaging-prediction model based on multimodal AI fusion algorithm.

Deep learning-based classification and mutation prediction from histopathological images of hepatocellular carcinoma.

Intervertebral disc degeneration (IDD) is a primary cause of chronic back pain and disability, highlighting the need for precise detection and grading for effective treatment. This study focuses on developing and validating a convolutional neural network (CNN) with a You Only Look Once (YOLO) architecture model using the Pfirrmann grading system to classify and grade lumbar intervertebral disc degeneration based on magnetic resonance imaging (MRI) scans. We developed a deep learning model trained on a dataset of anonymized MRI studies of patients with symptomatic back pain. MRI images were segmented and annotated by radiologists according to the Pfirrmann grading for the datasets. The segmentation MRI-disc image dataset was prepared for three groups: a training set (1,000), a testing set (500), and an external validation set (500) to assess model generalizability without overlapping images. The model's performance was evaluated using accuracy, sensitivity, specificity, F1 score, prediction error, and ROC-AUC. The AI model showed high performance across all metrics. For Grade I IDD, the model achieved an accuracy of 97%, 95%, and 92% in the training, testing, and external validation sets, respectively. For Grade II, the sensitivity was 100% in both training and testing sets and 98% in the validation set. For Grade III, the specificity was 95.4% in the training set and 94% in both testing and validation sets. For Grade IV, the F1 score was 97.77% in the training set and 95% in both testing and validation sets. For Grade V, the prediction error was 2.3%, 2%, and 2.5% in the training, testing, and validation sets, respectively. The overall ROC-AUC was 97%, 92%, and 95% in the training, testing, and validation sets, respectively. The AI-based classification model exhibits high accuracy, sensitivity, and specificity in detecting and grading lumbar IDD using the Pfirrmann grading. AI has significantly enhanced diagnostic precision and reliability, providing a powerful tool for clinicians in managing IDD. The potential impact is substantial, although further clinical validation is necessary before integrating this model into routine practice.

Predicting lymph node status based on deep learning on histopathological images from primary tumor of early-stage cervical squamous cell carcinoma (080)

ObjectiveTo examine the expression levels of miR-144-3p in the plasma of patients with gestational diabetes mellitus (GDM) and to construct a nomogram for predicting and evaluating factors influencing adverse pregnancy outcomes (APO) in GDM based on plasma miR-144-3p levels.MethodsThis study included 442 pregnant women, comprising 216 diagnosed with GDM (GDM group) and 226 with normal glucose tolerance (NGT group). Plasma miR-144-3p levels in both groups were measured using reverse transcription real-time polymerase chain reaction (RT-qPCR). The diagnostic performance of plasma miR-144-3p for GDM was assessed by receiver operating characteristic (ROC) curve analysis. During pregnancy, the GDM group was followed, and outcomes were categorized into two groups: 132 with favorable pregnancy outcomes (FPO) and 84 with APO. A random number table method was applied to divide the GDM group into a training set (n=151) and a validation set (n=65) using a 7:3 ratio. Differences in variables across pregnancy outcome subgroups in the training set were examined. Univariate and multivariate logistic regression analyses were performed to identify risk factors for APO in GDM. Based on these factors, a nomogram prediction model was developed to estimate the risk of APO in GDM. The model’s performance was evaluated using area under the curve (AUC) analysis, calibration curve analysis, and decision curve analysis (DCA).ResultsThe expression of miR-144-3p was significantly higher in the GDM group than in the NGT group (p < 0.05). miR-144-3p showed an AUC of 0.877, with a sensitivity of 81.09% and a specificity of 91.20% for diagnosing GDM. No statistically significant differences were observed in general clinical characteristics between the training and validation sets. In the training set, gestational weight gain (GWG), the number of OGTT abnormalities, glycaemic control (GC), and miR-144-3p expression varied significantly between the APO and FPO subgroups (p < 0.05). Multivariate logistic regression analysis identified increased GWG, the number of OGTT abnormalities, poor GC, and higher miR-144-3p levels as independent risk factors for APO in GDM. The AUC of the nomogram based on these variables was 0.881 in the training set and 0.855 in the validation set. Calibration curves indicated good agreement between predicted and actual outcomes in both sets. The DCA showed a clear net clinical benefit and stable predictive utility.ConclusionElevated plasma miR-144-3p levels in pregnant women with GDM may contribute to the occurrence of APO. The number of OGTT abnormalities and glycaemic control were also identified as independent risk factors. A nomogram incorporating miR-144-3p and these clinical indicators displays strong predictive accuracy and provides a practical tool for assessing APO risk in GDM.

With the global rise in advanced maternal age (AMA) pregnancies, the risk of gestational diabetes mellitus (GDM) increases. However, few GDM prediction models are tailored for AMA women. This study aims to develop a practical risk prediction model for GDM in AMA women. Data were obtained from a prospective observational cohort of AMA pregnant women from the Obstetrics and Gynecology Hospital in Shanghai, China. Singleton pregnancies with complete OGTT results at 24-28 weeks were selected and divided into training (70%) and validation (30%) sets. First-trimester predictors, including demographic, metabolic parameters, and clinical history, were evaluated for statistical significance. A multivariate logistic regression model was developed, with performance evaluated using receiver operating characteristic (ROC) curves and calibration plots. Predictors were primarily incorporated as categorical variables in a nomogram to enhance model convenience. A model using continuous predictors was also tested for comparison. A total of 1904 AMA women were included, with GDM incidence rates of 18.3% (243/1333) in the training set and 19.3% (110/571) in the validation set. Significant predictors for GDM diagnosis at 24-28 weeks included maternal age, GDM history, first-trimester fasting plasma glucose, mean arterial pressure, and triglyceride levels. The categorical model achieved an area under the ROC curve of 0.717 (95% CI: 0.682-0.753) in the training set and 0.702 (95% CI: 0.645-0.758) in the validation set. The Hosmer-Lemeshow test indicated good calibration (p = .97 in the training set; p = .66 in the validation set). The model with category and continuous predictors exhibited similar performance. This study developed and validated a practical early risk prediction nomogram for GDM in AMA women, using commonly available clinical data. The model shows good predictive performance and is resource-efficient, making it suitable for real-world clinical implementation.

A deep learning model for distinguishing pseudoprogression and tumor progression in glioblastoma based on pre- and post-operative contrast-enhanced T1 imaging.

The present study aimed to develop and validate a fusion model based on multi-phase contrast-enhanced computed tomography (CECT) radiomics features combined with clinical features to preoperatively predict the expression levels of Ki-67 in patients with gastric cancer (GC). A total of 164 patients with GC who underwent surgical treatment at our hospital between September 2015 and September 2023 were retrospectively included and were randomly divided into a training set (n=114) and a testing set (n=50). Using Pyradiomics, radiomics features were extracted from multi-phase CECT images and were combined with significant clinical features through various machine learning algorithms [support vector machine (SVM), random forest (RandomForest), K-nearest neighbors (KNN), LightGBM and XGBoost] to build a fusion model. Receiver operating characteristic, area under the curve (AUC), calibration curve and decision curve analysis (DCA) were used to evaluate, validate and compare the predictive performance and clinical utility of the model. Among the three single-phase models, for the arterial phase model, the SVM radiomics model had the highest AUC value in the training set, which was 0.697; and the RandomForest radiomics model had the highest AUC value in the testing set, which was 0.658. For the venous phase model, the SVM radiomics model had the highest AUC value in the training set, which was 0.783; and the LightGBM radiomics model had the highest AUC value in the testing set, which was 0.747. For the delayed phase model, the KNN radiomics model had the highest AUC value in the training set, which was 0.772; and the SVM radiomics model had the highest AUC in the testing set, which was 0.719. The clinical feature model had the lowest AUC values in both the training set and the testing set, which were 0.614 and 0.520, respectively. Notably, the multi-phase model and the fusion model, which were constructed by combining the clinical features and the multi-phase features, demonstrated excellent discriminative performance, with the fusion model achieving AUC values of 0.933 and 0.817 in the training and testing sets, thus outperforming other models (DeLong test, both P<0.05). The calibration curve showed that the fusion model had goodness of fit (Hosmer-Lemeshow test, >0.5 in the training and validation sets). The DCA showed that the net benefit of the fusion model in identifying high expression of Ki-67 was improved compared with that of other models. Furthermore, the fusion model achieved an AUC value of 0.805 in the external validation data from The Cancer Imaging Archive. In conclusion, the fusion model established in the present study was revealed to have excellent performance and is expected to serve as a non-invasive tool for predicting Ki-67 status and guiding clinical treatment.

Detection of fetal brain abnormalities using data augmentation and convolutional neural network in internet of things

Development of an Artificial Intelligence Model to Aid in Measurement of Invasion, Comprehensive Histologic Subtyping, and Grading of Pulmonary Adenocarcinoma.

Deep Convolutional Neural Network-Aided Detection of Portal Hypertension in Patients With Cirrhosis

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 develop and validate an automated magnetic resonance imaging (MRI)-based model to preoperatively differentiate pancreatic adenosquamous carcinoma (PASC) from pancreatic ductal adenocarcinoma (PDAC). This retrospective study included patients with surgically resected, histopathologically confirmed PASC or PDAC who underwent MRI between January 2011 and December 2020. According to time of treatment, they were divided into training and validation sets. Automated deep-learning-based artificial intelligence was used for pancreatic tumor segmentation. Linear discriminant analysis was performed with conventional MRI and radiomic features to develop clinical, radiomics, and mixed models in the training set. The models' performances were determined from their discrimination and clinical utility. Kaplan-Meier and log-rank tests were used for survival analysis. Overall, 389 and 123 patients with PDAC (age, 61.37 ± 9.47years; 251 men) and PASC (age, 61.99 ± 9.82years; 78 men) were included, respectively; they were split into the training (n = 358) and validation (n = 154) sets. The mixed model showed good performance in the training and validation sets (area under the curve: 0.94 and 0.96, respectively). The sensitivity, specificity, and accuracy were 76.74%, 93.38%, and 89.39% for the training set, respectively, and 67.57%, 97.44%, and 90.26% for the validation set, respectively. The mixed model outperformed the clinical (p = 0.001) and radiomics (p = 0.04) models in the validation set. Log-rank test revealed significantly longer survival in the predicted PDAC group than in the predicted PASC group (p = 0.003), according to the mixed model. Our mixed model, which combined MRI and radiomic features, can be used to differentiate PASC from PDAC.

ObjectiveThe risk category of gastric gastrointestinal stromal tumors (GISTs) are closely related to the surgical method, the scope of resection, and the need for preoperative chemotherapy. We aimed to develop and validate convolutional neural network (CNN) models based on preoperative venous-phase CT images to predict the risk category of gastric GISTs.MethodA total of 425 patients pathologically diagnosed with gastric GISTs at the authors’ medical centers between January 2012 and July 2021 were split into a training set (154, 84, and 59 with very low/low, intermediate, and high-risk, respectively) and a validation set (67, 35, and 26, respectively). Three CNN models were constructed by obtaining the upper and lower 1, 4, and 7 layers of the maximum tumour mask slice based on venous-phase CT Images and models of CNN_layer3, CNN_layer9, and CNN_layer15 established, respectively. The area under the receiver operating characteristics curve (AUROC) and the Obuchowski index were calculated to compare the diagnostic performance of the CNN models.ResultsIn the validation set, CNN_layer3, CNN_layer9, and CNN_layer15 had AUROCs of 0.89, 0.90, and 0.90, respectively, for low-risk gastric GISTs; 0.82, 0.83, and 0.83 for intermediate-risk gastric GISTs; and 0.86, 0.86, and 0.85 for high-risk gastric GISTs. In the validation dataset, CNN_layer3 (Obuchowski index, 0.871) provided similar performance than CNN_layer9 and CNN_layer15 (Obuchowski index, 0.875 and 0.873, respectively) in prediction of the gastric GIST risk category (All P >.05).ConclusionsThe CNN based on preoperative venous-phase CT images showed good performance for predicting the risk category of gastric GISTs.

ObjectivesPhosphatase and tensin homolog (PTEN) mutation is an indicator of poor prognosis of low-grade and high-grade glioma. This study built a reliable model from multi-parametric magnetic resonance imaging (MRI) for predicting the PTEN mutation status in patients with glioma.MethodsIn this study, a total of 244 patients with glioma were retrospectively collected from our center (n = 77) and The Cancer Imaging Archive (n = 167). All patients were randomly divided into a training set (n = 170) and a validation set (n = 74). Three models were built from preoperative MRI for predicting PTEN status, including a radiomics model, a convolutional neural network (CNN) model, and an integrated model based on both radiomics and CNN features. The performance of each model was evaluated by accuracy and the area under the receiver operating characteristic curve (AUC).ResultsThe CNN model achieved an AUC of 0.84 and an accuracy of 0.81, which performed better than did the radiomics model, with an AUC of 0.83 and an accuracy of 0.66. Combining radiomics with CNN will further benefit the predictive performance (accuracy = 0.86, AUC = 0.91).ConclusionsThe combination of both the CNN and radiomics features achieved significantly higher performance in predicting the mutation status of PTEN in patients with glioma than did the radiomics or the CNN model alone.

Identification of slightly sprouted wheat kernels using hyperspectral imaging technology and different deep convolutional neural networks