Our primary aim with this study was to build a patient-level classifier for stroke territory in DWI using AI to facilitate fast triage of stroke to a dedicated stroke center. A retrospective collection of DWI images of 271 and 122 consecutive acute ischemic stroke patients from two centers was carried out. Pretrained MobileNetV2 and EfficientNetB0 architectures were used to classify territorial subtypes as middle cerebral artery, posterior circulation, or watershed infarcts along with normal slices. Various input combinations using edge maps, thresholding, and hard attention versions were explored. The effect of augmenting the three-channel inputs of pre-trained models on classification performance was analyzed. ROC analyses and confusion matrix-derived performance metrics of the models were reported. Of the 271 patients included in this study, 151 (55.7%) were male and 120 (44.3%) were female. One hundred twenty-nine patients had MCA (47.6%), 65 patients had posterior circulation (24%), and 77 patients had watershed (28.0%) infarcts for center 1. Of the 122 patients from center 2, 78 (64%) were male and 44 (34%) were female. Fifty-two patients (43%) had MCA, 51 patients had posterior circulation (42%), and 19 (15%) patients had watershed infarcts. The Mobile-Crop model had the best performance with 0.95 accuracy and a 0.91 mean f1 score for slice-wise classification and 0.88 accuracy on external test sets, along with a 0.92 mean AUC. In conclusion, modified pre-trained models may be augmented with the transformation of images to provide a more accurate classification of affected territory by stroke in DWI.

With the increasing prevalence of mental illness, accurate clinical diagnosis of mental illness is crucial. Compared with MRI, CT has the advantages of wide application, low price, short scanning time, and high patient cooperation. This study aims to construct a deep learning (DL) model based on CT images to make identification of bipolar disorder (BD) and schizophrenia (SZ). A total of 506 patients (BD = 227, SZ = 279) and 179 healthy controls (HC) was collected from January 2022 to May 2023 at two hospitals, and divided into an internal training set and an internal validation set according to a ratio of 4:1. An additional 65 patients (BD = 35, SZ = 30) and 40 HC were recruited from different hospitals, and served as an external test set. All subjects accepted the conventional brain CT examination. The DenseMD model for identify BD and SZ using multiple instance learning was developed and compared with other classical DL models. The results showed that DenseMD performed excellently with an accuracy of 0.745 in the internal validation set, whereas the accuracy of the ResNet-18, ResNeXt-50, and DenseNet-121model was 0.672, 0.664, and 0.679, respectively. For the external test set, DenseMD again outperformed other models with an accuracy of 0.724; however, the accuracy of the ResNet-18, ResNeXt-50, and DenseNet-121model was 0.657, 0.638, and 0.676, respectively. Therefore, the potential of DL models for identification of BD and SZ based on brain CT images was established, and identification ability of the DenseMD model was better than other classical DL models.

Ultrasound-guided quadratus lumborum block (QLB) technology has become a widely used perioperative analgesia method during abdominal and pelvic surgeries. Due to the anatomical complexity and individual variability of the quadratus lumborum muscle (QLM) on ultrasound images, nerve blocks heavily rely on anesthesiologist experience. Therefore, using artificial intelligence (AI) to identify different tissue regions in ultrasound images is crucial. In our study, we retrospectively collected 112 patients (3162 images) and developed a deep learning model named Q-VUM, which is a U-shaped network based on the Visual Geometry Group 16 (VGG16) network. Q-VUM precisely segments various tissues, including the QLM, the external oblique muscle, the internal oblique muscle, the transversus abdominis muscle (collectively referred to as the EIT), and the bones. Furthermore, we evaluated Q-VUM. Our model demonstrated robust performance, achieving mean intersection over union (mIoU), mean pixel accuracy, dice coefficient, and accuracy values of 0.734, 0.829, 0.841, and 0.944, respectively. The IoU, recall, precision, and dice coefficient achieved for the QLM were 0.711, 0.813, 0.850, and 0.831, respectively. Additionally, the Q-VUM predictions showed that 85% of the pixels in the blocked area fell within the actual blocked area. Finally, our model exhibited stronger segmentation performance than did the common deep learning segmentation networks (0.734 vs. 0.720 and 0.720, respectively). In summary, we proposed a model named Q-VUM that can accurately identify the anatomical structure of the quadratus lumborum in real time. This model aids anesthesiologists in precisely locating the nerve block site, thereby reducing potential complications and enhancing the effectiveness of nerve block procedures.

This study developed and validated a deep learning-based diagnostic model with uncertainty estimation to aid radiologists in the preoperative differentiation of pathological subtypes of renal cell carcinoma (RCC) based on computed tomography (CT) images. Data from 668 consecutive patients with pathologically confirmed RCC were retrospectively collected from Center 1, and the model was trained using fivefold cross-validation to classify RCC subtypes into clear cell RCC (ccRCC), papillary RCC (pRCC), and chromophobe RCC (chRCC). An external validation with 78 patients from Center 2 was conducted to evaluate the performance of the model. In the fivefold cross-validation, the area under the receiver operating characteristic curve (AUC) for the classification of ccRCC, pRCC, and chRCC was 0.868 (95% CI, 0.826-0.923), 0.846 (95% CI, 0.812-0.886), and 0.839 (95% CI, 0.802-0.88), respectively. In the external validation set, the AUCs were 0.856 (95% CI, 0.838-0.882), 0.787 (95% CI, 0.757-0.818), and 0.793 (95% CI, 0.758-0.831) for ccRCC, pRCC, and chRCC, respectively. The model demonstrated robust performance in predicting the pathological subtypes of RCC, while the incorporated uncertainty emphasized the importance of understanding model confidence. The proposed approach, integrated with uncertainty estimation, offers clinicians a dual advantage: accurate RCC subtype predictions complemented by diagnostic confidence metrics, thereby promoting informed decision-making for patients with RCC.

This work aims to perform a cross-site validation of automated segmentation for breast cancers in MRI and to compare the performance to radiologists. A three-dimensional (3D) U-Net was trained to segment cancers in dynamic contrast-enhanced axial MRIs using a large dataset from Site 1 (n = 15,266; 449 malignant and 14,817 benign). Performance was validated on site-specific test data from this and two additional sites, and common publicly available testing data. Four radiologists from each of the three clinical sites provided two-dimensional (2D) segmentations as ground truth. Segmentation performance did not differ between the network and radiologists on the test data from Sites 1 and 2 or the common public data (median Dice score Site 1, network 0.86 vs. radiologist 0.85, n = 114; Site 2, 0.91 vs. 0.91, n = 50; common: 0.93 vs. 0.90). For Site 3, an affine input layer was fine-tuned using segmentation labels, resulting in comparable performance between the network and radiologist (0.88 vs. 0.89, n = 42). Radiologist performance differed on the common test data, and the network numerically outperformed 11 of the 12 radiologists (median Dice: 0.85-0.94, n = 20). In conclusion, a deep network with a novel supervised harmonization technique matches radiologists' performance in MRI tumor segmentation across clinical sites. We make code and weights publicly available to promote reproducible AI in radiology.

Techniques are developed for generating uncertainty estimates for convolutional neural network (CNN)-based methods for registering the locations of lesions between the craniocaudal (CC) and mediolateral oblique (MLO) mammographic X-ray image views. Multi-view lesion correspondence is an important task that clinicians perform for characterizing lesions during routine mammographic exams. Automated registration tools can aid in this task, yet if the tools also provide confidence estimates, they can be of greater value to clinicians, especially in cases involving dense tissue where lesions may be difficult to see. A set of deep ensemble-based techniques, which leverage a negative log-likelihood (NLL)-based cost function, are implemented for estimating uncertainties. The ensemble architectures involve significant modifications to an existing CNN dual-view lesion registration algorithm. Three architectural designs are evaluated, and different ensemble sizes are compared using various performance metrics. The techniques are tested on synthetic X-ray data, real 2D X-ray data, and slices from real 3D X-ray data. The ensembles generate covariance-based uncertainty ellipses that are correlated with registration accuracy, such that the ellipse sizes can give a clinician an indication of confidence in the mapping between the CC and MLO views. The results also show that the ellipse sizes can aid in improving computer-aided detection (CAD) results by matching CC/MLO lesion detects and reducing false alarms from both views, adding to clinical utility. The uncertainty estimation techniques show promise as a means for aiding clinicians in confidently establishing multi-view lesion correspondence, thereby improving diagnostic capability.

To compare the image quality and fat attenuation index (FAI) of coronary artery CT angiography (CCTA) under different tube voltages between deep learning image reconstruction (DLIR) and adaptive statistical iterative reconstruction V (ASIR-V). Three hundred one patients who underwent CCTA with automatic tube current modulation were prospectively enrolled and divided into two groups: 120 kV group and low tube voltage group. Images were reconstructed using ASIR-V level 50% (ASIR-V50%) and high-strength DLIR (DLIR-H). In the low tube voltage group, the voltage was selected according to Chinese BMI classification: 70 kV (BMI < 24 kg/m²), 80 kV (24 kg/m² ≤ BMI < 28 kg/m²), 100 kV (BMI ≥ 28 kg/m²). At the same tube voltage, the subjective and objective image quality, edge rise distance (ERD), and FAI between different algorithms were compared. Under different tube voltages, we used DLIR-H to compare the differences between subjective, objective image quality, and ERD. Compared with the 120 kV group, the DLIR-H image noise of 70 kV, 80 kV, and 100 kV groups increased by 36%, 25%, and 12%, respectively (all P < 0.001); contrast-to-noise ratio (CNR), subjective score, and ERD were similar (all P > 0.05). In the 70 kV, 80 kV, 100 kV, and 120 kV groups, compared with ASIR-V50%, DLIR-H image noise decreased by 50%, 53%, 47%, and 38-50%, respectively; CNR, subjective score, and FAI value increased significantly (all P < 0.001), ERD decreased. Compared with 120 kV tube voltage, the combination of DLIR-H and low tube voltage maintains image quality. At the same tube voltage, compared with ASIR-V, DLIR-H improves image quality and FAI value.

To establish a reliable machine learning model to predict malignancy in breast lesions identified by ultrasound (US) and optimize the negative predictive value to minimize unnecessary biopsies. We included clinical and ultrasonographic attributes from 1526 breast lesions classified as BI-RADS 3, 4a, 4b, 4c, 5, and 6 that underwent US-guided breast biopsy in four institutions. We selected the most informative attributes to train nine machine learning models, ensemble models and models with tuned threshold to make inferences about the diagnosis of BI-RADS 4a and 4b lesions (validation dataset). We tested the performance of the final model with 403 new suspicious lesions. The most informative attributes were shape, margin, orientation and size of the lesions, the resistance index of the internal vessel, the age of the patient and the presence of a palpable lump. The highest mean negative predictive value (NPV) was achieved with the K-Nearest Neighbors algorithm (97.9%). Making ensembles did not improve the performance. Tuning the threshold did improve the performance of the models and we chose the algorithm XGBoost with the tuned threshold as the final one. The tested performance of the final model was: NPV 98.1%, false negative 1.9%, positive predictive value 77.1%, false positive 22.9%. Applying this final model, we would have missed 2 of the 231 malignant lesions of the test dataset (0.8%). Machine learning can help physicians predict malignancy in suspicious breast lesions identified by the US. Our final model would be able to avoid 60.4% of the biopsies in benign lesions missing less than 1% of the cancer cases.

The purpose of this study is to assess segmentation reproducibility of artificial intelligence-based algorithm, TotalSegmentator, across 34 anatomical structures using multiphasic abdominal CT scans comparing unenhanced, arterial, and portal venous phases in the same patients. A total of 1252 multiphasic abdominal CT scans acquired at our institution between January 1, 2012, and December 31, 2022, were retrospectively included. TotalSegmentator was used to derive volumetric measurements of 34 abdominal organs and structures from the total of 3756 CT series. Reproducibility was evaluated across three contrast phases per CT and compared to two human readers and an independent nnU-Net trained on the BTCV dataset. Relative deviation in segmented volumes and absolute volume deviations (AVD) were reported. Volume deviation within 5% was considered reproducible. Thus, non-inferiority testing was conducted using a 5% margin. Twenty-nine out of 34 structures had volume deviations within 5% and were considered reproducible. Volume deviations for the adrenal glands, gallbladder, spleen, and duodenum were above 5%. Highest reproducibility was observed for bones (- 0.58% [95% CI: - 0.58, - 0.57]) and muscles (- 0.33% [- 0.35, - 0.32]). Among abdominal organs, volume deviation was 1.67% (1.60, 1.74). TotalSegmentator outperformed the reproducibility of the nnU-Net trained on the BTCV dataset with an AVD of 6.50% (6.41, 6.59) vs. 10.03% (9.86, 10.20; p < 0.0001), most notably in cases with pathologic findings. Similarly, TotalSegmentator's AVD between different contrast phases was superior compared to the interreader AVD for the same contrast phase (p = 0.036). TotalSegmentator demonstrated high intra-individual reproducibility for most abdominal structures in multiphasic abdominal CT scans. Although reproducibility was lower in pathologic cases, it outperforms both human readers and a nnU-Net trained on the BTCV dataset.

Malignant cerebral edema (MCE) is a significant complication following endovascular thrombectomy (EVT) in the treatment of acute ischemic stroke. This study aimed to develop and validate a deep learning-assisted diagnosis model based on the hyperattenuated imaging marker (HIM), characterized by hyperattenuation on head non-contrast computed tomography immediately after thrombectomy, to facilitate radiologists in predicting MCE in patients receiving EVT. This study included 271 patients, with 168 in the training cohort, 43 in the validation cohort, and 60 in the prospective internal test cohort. Deep learning models including ResNet 50, ResNet 101, ResNeXt50_32×4d, ResNeXt101_32×8d, and DenseNet 121 were constructed. The performance of senior and junior radiologists with and without optimal model assistance was compared. ResNeXt101_32×8d had the best predictive performance, the analysis of the receiver operating characteristic curve indicated an area under the curve (AUC) of 0.897 for the prediction of MCE in the validation group and an AUC of 0.889 in the test group. Moreover, with the assistance of the model, radiologists exhibited a significant improvement in diagnostic performance, the AUC increased by 0.137 for the junior radiologist and 0.096 for the junior radiologist respectively. Our study utilized the ResNeXt-101 neural network, combined with HIM, to validate a deep learning model for predicting MCE post-EVT. The developed deep learning model demonstrated high discriminative ability, and can serve as a valuable adjunct to radiologists in clinical practice.