This study aimed to predict and classify MRI PI-RADs scores using different machine learning algorithms and to detect the concordance of PI-RADs scoring with the outcome target of prostate biopsy. Machine learning (ML) algorithms were used to develop best-fitting models for the prediction and classification of MRI PI-RAD. The Random Forest and Extra Trees models achieved the best performance compared to the other methods. The accuracy of both models was 91.95%. The AUC was 0.9329 for the Random Forest model and 0.9404 for the Extra Trees model. PSA level, PSA density, and diameter of the largest lesion were the most important features for the importance of outcome classification. ML prediction enhanced the PI-RAD classification, where clinically significant prostate cancer (csPCa) cases increased from 0% to 1.9% in the low-risk PI-RAD class, this showed that the model identified some previously missed cases. Predictive machine learning models showed an excellent ability to predict MRI Pi-RAD scores and discriminate between low- and high-risk scores. However, caution should be exercised, as a high percentage of negative biopsy cases were assigned Pi-RAD 4 and Pi-RAD 5 scores. ML integration may enhance PI-RAD's utility by reducing unnecessary biopsies in low-risk patients (via better csPCa detection) and refining the high-risk categorization. Combining such PI-RAD scores with significant parameters, such as PSA density, lesion diameter, number of lesions, and age, in decision curve analysis and utility paradigms would assist physicians' clinical decisions.

The automatic screening of thyroid nodules using computer-aided diagnosis holds great promise in reducing missed and misdiagnosed cases in clinical practice. However, most current research focuses on single-modal images and does not fully leverage the comprehensive information from multimodal medical images, limiting model performance. To enhance screening accuracy, this study uses a deep learning framework that integrates high-dimensional convolutions of B-mode ultrasound (BMUS) and strain elastography (SE) images to predict the malignancy of TI-RADS 4 thyroid nodules with high-risk features. First, we extract nodule regions from the images and expand the boundary areas. Then, adaptive particle swarm optimization (APSO) and contrast limited adaptive histogram equalization (CLAHE) algorithms are applied to enhance ultrasound image contrast. Finally, deep learning techniques are used to extract and fuse high-dimensional features from both ultrasound modalities to classify benign and malignant thyroid nodules. The proposed model achieved an AUC of 0.937 (95 % CI 0.917-0.949) and 0.927 (95 % CI 0.907-0.948) in the test and external validation sets, respectively, demonstrating strong generalization ability. When compared with the diagnostic performance of three groups of radiologists, the model outperformed them significantly. Meanwhile, with the model's assistance, all three radiologist groups showed improved diagnostic performance. Furthermore, heatmaps generated by the model show a high alignment with radiologists' expertise, further confirming its credibility. The results indicate that our model can assist in clinical thyroid nodule diagnosis, reducing the risk of missed and misdiagnosed diagnoses, particularly for high-risk populations, and holds significant clinical value.

MRI sequence classification becomes challenging in multicenter studies due to variability in imaging protocols, leading to unreliable metadata and requiring labor-intensive manual annotation. While numerous automated MRI sequence identification models are available, they frequently encounter the issue of domain shift, which detrimentally impacts their accuracy. This study addresses domain shift, particularly from adult to pediatric MRI data, by evaluating the effectiveness of pre-trained models under these conditions. This retrospective and multicentric study explored the efficiency of a pre-trained convolutional (ResNet) and CNN-Transformer hybrid model (MedViT) to handle domain shift. The study involved training ResNet-18 and MedVit models on an adult MRI dataset and testing them on a pediatric dataset, with expert domain knowledge adjustments applied to account for differences in sequence types. The MedViT model demonstrated superior performance compared to ResNet-18 and benchmark models, achieving an accuracy of 0.893 (95% CI 0.880-0.904). Expert domain knowledge adjustments further improved the MedViT model's accuracy to 0.905 (95% CI 0.893-0.916), showcasing its robustness in handling domain shift. Advanced neural network architectures like MedViT and expert domain knowledge on the target dataset significantly enhance the performance of MRI sequence classification models under domain shift conditions. By combining the strengths of CNNs and transformers, hybrid architectures offer enhanced robustness for reliable automated MRI sequence classification in diverse research and clinical settings. Question Domain shift between adult and pediatric MRI data limits deep learning model accuracy, requiring solutions for reliable sequence classification across diverse patient populations. Findings The MedViT model outperformed ResNet-18 in pediatric imaging; expert domain knowledge adjustment further improved accuracy, demonstrating robustness across diverse datasets. Clinical relevance This study enhances MRI sequence classification by leveraging advanced neural networks and expert domain knowledge to mitigate domain shift, boosting diagnostic precision and efficiency across diverse patient populations in multicenter environments.

Current study aimed to investigate radiomics features derived from two-center diffusion-MRI to differentiate benign and hepatocellular carcinoma (HCC) liver nodules. A total of 328 patients with 517 LI-RADS 2-5 nodules were included. MR images were retrospectively collected from 3 T and 1.5 T MRI vendors. Lesions were categorized into 242 benign and 275 HCC based on follow-up imaging for LR-2,3 and pathology results for LR4,5 nodules, and randomly divided into training (80%) and test (20%) sets. Preprocessing included resampling and normalization. Radiomics features were extracted from lesion volume-of-interest (VOI) on diffusion Images. Scanner variability was corrected using ComBat harmonization method followed by High-correlation filter, PCA filter, and LASSO to select important features. Best classifier model was selected by 10-fold cross-validation, and accuracy was assessed on the test dataset. 1,434 features were extracted, and subsequent classifiers were constructed based on the 16 most important selected features. Notably, support-vector machine (SVM) demonstrated better performance in the test dataset in distinguishing between benign and HCC nodules, achieving an accuracy of 0.92, sensitivity of 0.94, and specificity of 0.86. Utilizing diffusion-MRI radiomics, our study highlights the performance of SVM, trained on lesions' diffusivity characteristics, in distinguishing benign and HCC nodules, ensuring clinical potential. It is suggested that further evaluations be conducted on multi-center datasets to address harmonization challenges. Integration of diffusion radiomics, for monitoring water restriction patterns as tumor histopathological index, with machine learning models demonstrates potential for achieving a reliable noninvasive method to improve the current diagnosis criteria.

Brain tumors pose a severe health risk, often leading to fatal outcomes if not detected early. While most studies focus on improving classification accuracy, this research emphasizes prediction certainty, quantified through loss values. Traditional metrics like accuracy and precision do not capture confidence in predictions, which is critical for medical applications. This study establishes a correlation between lower loss values and higher prediction certainty, ensuring more reliable tumor classification. We evaluate CNN, ResNet50, XceptionNet, and a Proposed Model (VGG19 with customized classification layers) using accuracy, precision, recall, and loss. Results show that while accuracy remains comparable across models, the Proposed Model achieves the best performance (96.95 % accuracy, 0.087 loss), outperforming others in both precision and recall. These findings demonstrate that certainty-aware AI models are essential for reliable clinical decision-making. This study highlights the potential of AI to bridge the shortage of medical professionals by integrating reliable diagnostic tools in healthcare. AI-powered systems can enhance early detection and improve patient outcomes, reinforcing the need for certainty-driven AI adoption in medical imaging.

Although there are notable structural abnormalities in the brain associated with psychotic diseases, it is still unclear how these abnormalities relate to clinical presentation. However, the fractal dimension (FD), which offers details on the complexity and irregularity of brain microstructures, may be a promising feature, as demonstrated by neuropsychiatric disorders such as Parkinson's and Alzheimer's. It may offer a possible biomarker for the detection and prognosis of psychosis when paired with machine learning. The purpose of this study is to investigate FD as a structural magnetic resonance imaging (sMRI) feature from individuals with a high clinical risk of psychosis who did not transit to psychosis (CHR_NT), clinical high risk who transit to psychosis (CHR_T), patients with first-episode psychosis (FEP) and healthy controls (HC). Using a machine learning approach that ultimately classifies sMRI images, the goals are (a) to evaluate FD as a potential biomarker and (b) to investigate its ability to predict a subsequent transition to psychosis from the high-risk clinical condition. We obtained sMRI images from 194 subjects, including 44 HCs, 77 FEPs, 16 CHR_Ts, and 57 CHR_NTs. We extracted the FD features and analyzed them using machine learning methods under five classification schemas (a) FEP vs. HC, (b) FEP vs. CHR_NT, (c) FEP vs. CHR_T, (d) CHR_NT vs. CHR_T, (d) CHR_NT vs. HC and (e) CHR_T vs. HC. In addition, the CHR_T group was used as external validation in (a), (b) and (d) comparisons to examine whether the progression of the disorder followed the FEP or CHR_NT patterns. The proposed algorithm resulted in a balanced accuracy greater than 0.77. This study has shown that FD can function as a predictive neuroimaging marker, providing fresh information on the microstructural alterations triggered throughout the course of psychosis. The effectiveness of FD in the detection of psychosis and transition to psychosis should be established by further research using larger datasets.

Accurate patient dosimetry estimates from fluoroscopically-guided interventions (FGIs) are hindered by limited knowledge of the specific anatomy that was irradiated. Current methods use data reported by the equipment to estimate the patient anatomy exposed during each irradiation event. We propose a deep learning algorithm to automatically match 2D fluoroscopic images with corresponding anatomical regions in computational phantoms, enabling more precise patient dose estimates. Our method involves two main steps: (1) simulating 2D fluoroscopic images, and (2) developing a deep learning algorithm to predict anatomical coordinates from these images. For part (1), we utilized DeepDRR for fast and realistic simulation of 2D x-ray images from 3D computed tomography datasets. We generated a diverse set of simulated fluoroscopic images from various regions with different field sizes. In part (2), we employed a Residual Neural Network (ResNet) architecture combined with metadata processing to effectively integrate patient-specific information (age and gender) to learn the transformation between 2D images and specific anatomical coordinates in each representative phantom. For the Modified ResNet model, we defined an allowable error range of ± 10 mm. The proposed method achieved over 90% of predictions within ± 10 mm, with strong alignment between predicted and true coordinates as confirmed by Bland-Altman analysis. Most errors were within ± 2%, with outliers beyond ± 5% primarily in Z-coordinates for infant phantoms due to their limited representation in the training data. These findings highlight the model's accuracy and its potential for precise spatial localization, while emphasizing the need for improved performance in specific anatomical regions. In this work, a comprehensive simulated 2D fluoroscopy image dataset was developed, addressing the scarcity of real clinical datasets and enabling effective training of deep-learning models. The modified ResNet successfully achieved precise prediction of anatomical coordinates from the simulated fluoroscopic images, enabling the goal of more accurate patient-specific dosimetry.

To develop and validate a machine learning model based on dual-energy computed tomography (DECT) for predicting cervical lymph node metastases (CLNM) in patients diagnosed with nasopharyngeal carcinoma (NPC). This prospective single-center study enrolled patients with NPC and the study assessment included both DECT and 18F-fluorodeoxyglucose positron emission tomography/computed tomography (18F-FDG PET/CT). Radiomics features were extracted from each region of interest (ROI) for cervical lymph nodes using arterial and venous phase images at 100 keV and 150 keV, either individually as non-fusion models or combined as fusion models on the DECT images. The performance of the random forest (RF) models, combined with radiomics features, was evaluated by area under the receiver operating characteristic curve (AUC) analysis. DeLong's test was employed to compare model performances, while decision curve analysis (DCA) assessed the clinical utility of the predictive models. Sixty-six patients with NPC were included for analysis, which was divided into a training set (n = 42) and a validation set (n = 22). A total of 13 radiomic models were constructed (4 non-fusion models and 9 fusion models). In the non-fusion models, when the threshold value exceeded 0.4, the venous phase at 100 keV (V100) (AUC, 0.9667; 95 % confidence interval [95 % CI], 0.9363-0.9901) model exhibited a higher net benefit than other non-fusion models. The V100 + V150 fusion model achieved the best performance, with an AUC of 0.9697 (95 % CI, 0.9393-0.9907). DECT-based radiomics effectively diagnosed CLNM in patients with NPC and may potentially be a valuable tool for clinical decision-making. This study improved pre-operative evaluation, treatment strategy selection, and prognostic evaluation for patients with nasopharyngeal carcinoma by combining DECT and radiomics to predict cervical lymph node status prior to treatment.

Major Depressive Disorder (MDD) is a prevalent mental health condition with significant societal impact. Structural magnetic resonance imaging (sMRI) and machine learning have shown promise in psychiatry, offering insights into brain abnormalities in MDD. However, predicting treatment response remains challenging. This study leverages inter-brain similarity from sMRI as a novel feature to enhance prediction accuracy and explore disease mechanisms. The method's generalizability across adult and adolescent cohorts is also evaluated. The study included 172 participants. Based on remission status, 39 participants from the Hangzhou Dataset and 34 from the Jinan Dataset were selected for further analysis. Three methods were used to extract brain similarity features, followed by a statistical test for feature selection. Six machine learning classifiers were employed to predict treatment response, and their generalizability was tested using the Jinan Dataset. Group analyses between remission and non-remission groups were conducted to identify brain regions associated with treatment response. Brain similarity features outperformed traditional metrics in predicting treatment outcomes, with the highest accuracy achieved by the model using these features. Between-group analyses revealed that the remission group had lower gray matter volume and density in the right precentral gyrus, but higher white matter volume (WMV). In the Jinan Dataset, significant differences were observed in the right cerebellum and fusiform gyrus, with higher WMV and density in the remission group. This study demonstrates that brain similarity features combined with machine learning can predict treatment response in MDD with moderate success across age groups. These findings emphasize the importance of considering age-related differences in treatment planning to personalize care. Clinical trial number: not applicable.

Since coronary artery disease (CAD) is a common comorbidity in patients with aortic valve stenosis, invasive coronary angiography (ICA) can be avoided if significant CAD can be screened with the non-invasive coronary CT angiography (cCTA). This study aims to evaluate the ability of machine learning-based CT coronary fractional flow reserve (CT-FFR) derived from cCTA to aid in the diagnosis of comorbid CAD in patients undergoing transcatheter aortic valve implantation (TAVI). A total of 100 patients who underwent both cCTA and ICA assessments prior to TAVI procedure between January 2021 and July 2023 were included. Coronary stenosis was assessed using both cCTA data and machine learning-generated CT-FFR image information for patients/major coronary vessels. Coronary lesions with CT-FFR ≤ 0.80 were defined as hemodynamically significant, with ICA serving as the diagnostic gold standard. A total of 400 major coronary vessels were identified in 100 eligible patients who underwent TAVI. CT-FFR was 86.4% sensitive and 66.1% specific to diagnose CAD, with a positive predictive value (PPV) of 66.7% and a negative predictive value (NPV) of 86.0%. The diagnostic accuracy (Acc) was 75.0%, with a false positive rate (FPR) of 33.9%. At the vessel level, CT-FFR showed a sensitivity of 77.6% and a specificity of 76.9%. The PPV was 44.0% and the NPV was 93.6%. The Acc was 77.0% and the FPR was 23.1%. For all patient/vessel units, CT-FFR outperformed cCTA. Machine learning-based CT-FFR can effectively detect coronary hemodynamic abnormalities. Combined with preoperative cCTA in TAVI patients, it is an effective tool to rule out significant CAD, reducing unnecessary coronary angiography in this high-risk population. Not applicable.