Breast cancer risk depends on an accurate assessment of breast density due to lesion masking. Although governed by standardized guidelines, radiologist assessment of breast density is still highly variable. Automated breast density assessment tools leverage deep learning but are limited by model robustness and interpretability. We assessed the robustness of a feature selection methodology (RFE-SHAP) for classifying breast density grades using tissue-specific radiomic features extracted from raw central projections of digital breast tomosynthesis screenings ( <math xmlns="http://www.w3.org/1998/Math/MathML"> <mrow> <msub><mrow><mi>n</mi></mrow> <mrow><mi>I</mi></mrow> </msub> <mo>=</mo> <mn>651</mn></mrow> </math> , <math xmlns="http://www.w3.org/1998/Math/MathML"> <mrow> <msub><mrow><mi>n</mi></mrow> <mrow><mi>II</mi></mrow> </msub> <mo>=</mo> <mn>100</mn></mrow> </math> ). RFE-SHAP leverages traditional and explainable AI methods to identify highly predictive and influential features. A simple logistic regression (LR) classifier was used to assess classification performance, and unsupervised clustering was employed to investigate the intrinsic separability of density grade classes. LR classifiers yielded cross-validated areas under the receiver operating characteristic (AUCs) per density grade of [ <math xmlns="http://www.w3.org/1998/Math/MathML"><mrow><mi>A</mi></mrow> </math> : <math xmlns="http://www.w3.org/1998/Math/MathML"><mrow><mn>0.909</mn> <mo>±</mo> <mn>0.032</mn></mrow> </math> , <math xmlns="http://www.w3.org/1998/Math/MathML"><mrow><mi>B</mi></mrow> </math> : <math xmlns="http://www.w3.org/1998/Math/MathML"><mrow><mn>0.858</mn> <mo>±</mo> <mn>0.027</mn></mrow> </math> , <math xmlns="http://www.w3.org/1998/Math/MathML"><mrow><mi>C</mi></mrow> </math> : <math xmlns="http://www.w3.org/1998/Math/MathML"><mrow><mn>0.927</mn> <mo>±</mo> <mn>0.013</mn></mrow> </math> , <math xmlns="http://www.w3.org/1998/Math/MathML"><mrow><mi>D</mi></mrow> </math> : <math xmlns="http://www.w3.org/1998/Math/MathML"><mrow><mn>0.890</mn> <mo>±</mo> <mn>0.089</mn></mrow> </math> ] and an AUC of <math xmlns="http://www.w3.org/1998/Math/MathML"><mrow><mn>0.936</mn> <mo>±</mo> <mn>0.016</mn></mrow> </math> for classifying patients as nondense or dense. In external validation, we observed per density grade AUCs of [ <math xmlns="http://www.w3.org/1998/Math/MathML"><mrow><mi>A</mi></mrow> </math> : 0.880, <math xmlns="http://www.w3.org/1998/Math/MathML"><mrow><mi>B</mi></mrow> </math> : 0.779, <math xmlns="http://www.w3.org/1998/Math/MathML"><mrow><mi>C</mi></mrow> </math> : 0.878, <math xmlns="http://www.w3.org/1998/Math/MathML"><mrow><mi>D</mi></mrow> </math> : 0.673] and nondense/dense AUC of 0.823. Unsupervised clustering highlighted the ability of these features to characterize different density grades. Our RFE-SHAP feature selection methodology for classifying breast tissue density generalized well to validation datasets after accounting for natural class imbalance, and the identified radiomic features properly captured the progression of density grades. Our results potentiate future research into correlating selected radiomic features with clinical descriptors of breast tissue density.

Explainable Artificial Intelligence (XAI) encompasses a broad spectrum of methods that aim to enhance the transparency of deep learning models, with Class Activation Mapping (CAM) methods widely used for visual interpretability. However, systematic evaluations of these methods in veterinary radiography remain scarce. This study presents a comparative analysis of eleven CAM methods, including GradCAM, XGradCAM, ScoreCAM, and EigenCAM, on a dataset of 7362 canine and feline X-ray images. A ResNet18 model was chosen based on the specificity of the dataset and preliminary results where it outperformed other models. Quantitative and qualitative evaluations were performed to determine how well each CAM method produced interpretable heatmaps relevant to clinical decision-making. Among the techniques evaluated, EigenGradCAM achieved the highest mean score and standard deviation (SD) of 2.571 (SD = 1.256), closely followed by EigenCAM at 2.519 (SD = 1.228) and GradCAM++ at 2.512 (SD = 1.277), with methods such as FullGrad and XGradCAM achieving worst scores of 2.000 (SD = 1.300) and 1.858 (SD = 1.198) respectively. Despite variations in saliency visualization, no single method universally improved veterinarians' diagnostic confidence. While certain CAM methods provide better visual cues for some pathologies, they generally offered limited explainability and didn't substantially improve veterinarians' diagnostic confidence.

To evaluate the post-deployment performance of an artificial intelligence (AI) system (Aidoc) for intracranial hemorrhage (ICH) detection and assess the utility of ChatGPT-4 Turbo for automated AI monitoring. This retrospective study evaluated 332,809 head CT examinations from 37 radiology practices across the United States (December 2023-May 2024). Of these, 13,569 cases were flagged as positive for ICH by the Aidoc AI system. A HIPAA (Health Insurance Portability and Accountability Act) -compliant version of ChatGPT-4 Turbo was used to extract data from radiology reports. Ground truth was established through radiologists' review of 200 randomly selected cases. Performance metrics were calculated for ChatGPT, Aidoc and radiologists. ChatGPT-4 Turbo demonstrated high diagnostic accuracy in identifying intracranial hemorrhage (ICH) from radiology reports, with a positive predictive value of 1 and a negative predictive value of 0.988 (AUC:0.996). Aidoc's false positive classifications were influenced by scanner manufacturer, midline shift, mass effect, artifacts, and neurologic symptoms. Multivariate analysis identified Philips scanners (OR: 6.97, p=0.003) and artifacts (OR: 3.79, p=0.029) as significant contributors to false positives, while midline shift (OR: 0.08, p=0.021) and mass effect (OR: 0.18, p=0.021) were associated with a reduced false positive rate. Aidoc-assisted radiologists achieved a sensitivity of 0.936 and a specificity of 1. This study underscores the importance of continuous performance monitoring for AI systems in clinical practice. The integration of LLMs offers a scalable solution for evaluating AI performance, ensuring reliable deployment and enhancing diagnostic workflows.

Clinical magnetic resonance imaging (MRI) is a high-resolution tool widely used for detailed anatomical imaging. However, prolonged scan times often lead to motion artefacts and patient discomfort. Fast acquisition techniques can reduce scan times but often produce noisy, low-contrast images, compromising segmentation accuracy essential for diagnosis and treatment planning. To address these limitations, we developed an end-to-end framework that incorporates BIDS-based data organiser and anonymizer, a GAN-based MR image enhancement model (GAN-MRI), AssemblyNet for brain region segmentation, and an attention-residual U-Net with Guided loss for abdominal and thigh segmentation. Thirty brain scans (5,400 slices) and 32 abdominal (1,920 slices) and 55 thigh scans (2,200 slices) acquired from multiple MRI scanners (GE, Siemens, Toshiba) underwent evaluation. Image quality improved significantly, with SNR and CNR for brain scans increasing from 28.44 to 42.92 (p < 0.001) and 11.88 to 18.03 (p < 0.001), respectively. Abdominal scans exhibited SNR increases from 35.30 to 50.24 (p < 0.001) and CNR from 10,290.93 to 93,767.22 (p < 0.001). Double-blind evaluations highlighted improved visualisations of anatomical structures and bias field correction. Segmentation performance improved substantially in the thigh (muscle: + 21%, IMAT: + 9%) and abdominal regions (SSAT: + 1%, DSAT: + 2%, VAT: + 12%), while brain segmentation metrics remained largely stable, reflecting the robustness of the baseline model. Proposed framework is designed to handle data from multiple anatomies with variations from different MRI scanners and centres by enhancing MRI scan and improving segmentation accuracy, diagnostic precision and treatment planning while reducing scan times and maintaining patient comfort.

Denoising reconstructed Computed Tomography (CT) images without access to raw projection data remains a significant difficulty in medical imaging, particularly when utilizing sharp or medium reconstruction kernels that generate high-frequency noise. This work introduces an innovative method that integrates quaternion mathematics with bilateral filtering to resolve this issue. The proposed Quaternion Bilateral Filter (QBF) effectively maintains anatomical structures and mitigates noise caused by the kernel by expressing CT scans in quaternion form, with the red, green, and blue channels encoded together. Compared to conventional methods that depend on raw data or grayscale filtering, our approach functions directly on reconstructed sharp kernel images. It converts them to mimic the quality of soft-kernel outputs, obtained with kernels such as B30f, using paired data from the same patients. The efficacy of the QBF is evidenced by both full-reference metrics (Structural Similarity Index Measure (SSIM), Peak Signal-to-Noise Ratio (PSNR), Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE)) and no-reference perceptual metrics (Naturalness Image Quality Evaluator (NIQE), Blind Referenceless Image Spatial Quality Evaluator (BRISQUE), and Perception-based Image Quality Evaluator (PIQE)). The results indicate that the QBF demonstrates improved denoising efficacy compared to traditional Bilateral Filter (BF), Non-Local Means (NLM), wavelet, and Convolutional Neural Network (CNN)-based processing, achieving an SSIM of 0.96 and a PSNR of 36.3 on B50f reconstructions. Additionally, segmentation-based visual validation verifies that QBF-filtered outputs maintain essential structural details necessary for subsequent diagnostic tasks. This study emphasizes the importance of quaternion-based filtering as a lightweight, interpretable, and efficient substitute for deep learning models in post-reconstruction CT image enhancement.

Accurate prediction of post-stroke language outcomes using machine learning offers the potential to enhance clinical treatment and rehabilitation for aphasic patients. This study of 758 English speaking stroke patients from the PLORAS project explores the impact of sample size on the performance of logistic regression and a deep learning (ResNet-18) model in predicting language outcomes from neuroimaging and impairment-relevant tabular data. We assessed the performance of both models on two key language tasks from the Comprehensive Aphasia Test: Spoken Picture Description and Naming, using a learning curve approach. Contrary to expectations, the simpler logistic regression model performed comparably or better than the deep learning model (with overlapping confidence intervals), with both models showing an accuracy plateau around 80% for sample sizes larger than 300 patients. Principal Component Analysis revealed that the dimensionality of the neuroimaging data could be reduced to as few as 20 (or even 2) dominant components without significant loss in accuracy, suggesting that classification may be driven by simple patterns such as lesion size. The study highlights both the potential limitations of current dataset size in achieving further accuracy gains and the need for larger datasets to capture more complex patterns, as some of our results indicate that we might not have reached an absolute classification performance ceiling. Overall, these findings provide insights into the practical use of machine learning for predicting aphasia outcomes and the potential benefits of much larger datasets in enhancing model performance.

As an important branch of machine learning pipelines in medical imaging, radiomics faces two major challenges namely reproducibility and accessibility. In this work, we introduce open-radiomics, a set of radiomics datasets along with a comprehensive radiomics pipeline based on our proposed technical protocol to investigate the effects of radiomics feature extraction on the reproducibility of the results. We curated large-scale radiomics datasets based on three open-source datasets; BraTS 2020 for high-grade glioma (HGG) versus low-grade glioma (LGG) classification and survival analysis, BraTS 2023 for O6-methylguanine-DNA methyltransferase (MGMT) classification, and non-small cell lung cancer (NSCLC) survival analysis from the Cancer Imaging Archive (TCIA). We used the BraTS 2020 open-source Magnetic Resonance Imaging (MRI) dataset to demonstrate how our proposed technical protocol could be utilized in radiomics-based studies. The cohort includes 369 adult patients with brain tumors (76 LGG, and 293 HGG). Using PyRadiomics library for LGG vs. HGG classification, we created 288 radiomics datasets; the combinations of 4 MRI sequences, 3 binWidths, 6 image normalization methods, and 4 tumor subregions. We used Random Forest classifiers, and for each radiomics dataset, we repeated the training-validation-test (60%/20%/20%) experiment with different data splits and model random states 100 times (28,800 test results) and calculated the Area Under the Receiver Operating Characteristic Curve (AUROC). Unlike binWidth and image normalization, the tumor subregion and imaging sequence significantly affected performance of the models. T1 contrast-enhanced sequence and the union of Necrotic and the non-enhancing tumor core subregions resulted in the highest AUROCs (average test AUROC 0.951, 95% confidence interval of (0.949, 0.952)). Although several settings and data splits (28 out of 28800) yielded test AUROC of 1, they were irreproducible. Our experiments demonstrate the sources of variability in radiomics pipelines (e.g., tumor subregion) can have a significant impact on the results, which may lead to superficial perfect performances that are irreproducible. Not applicable.

The Artificial Intelligence in Medical Imaging (AIMI) initiative aims to enhance the National Cancer Institute's (NCI) Image Data Commons (IDC) by releasing fully reproducible nnU-Net models, along with AI-assisted segmentation for cancer radiology images. In this extension of our earlier work, we created high-quality, AI-annotated imaging datasets for 11 IDC collections, spanning computed tomography (CT) and magnetic resonance imaging (MRI) of the lungs, breast, brain, kidneys, prostate, and liver. Each nnU-Net model was trained on open-source datasets, and a portion of the AI-generated annotations was reviewed and corrected by board-certified radiologists. Both the AI and radiologist annotations were encoded in compliance with the Digital Imaging and Communications in Medicine (DICOM) standard, ensuring seamless integration into the IDC collections. By making these models, images, and annotations publicly accessible, we aim to facilitate further research and development in cancer imaging.

Diagnosis of prostate cancer requires histopathology of tissue samples. Following an MRI to identify suspicious areas, a biopsy is performed under ultrasound (US) guidance. In existing assistance systems, 3D US information is generally available (taken before the biopsy session and/or in between samplings). However, without registration between 2D images and 3D volumes, the urologist must rely on cognitive navigation. This work introduces a deep learning model to track the orientation of real-time US slices relative to a reference 3D US volume using only image and volume data. The dataset comprises 515 3D US volumes collected from 51 patients during routine transperineal biopsy. To generate 2D images streams, volumes are resampled to simulate three degrees of freedom rotational movements around the rectal entrance. The proposed model comprises two ResNet-based sub-modules to address the symmetry ambiguity arising from complex out-of-plane movement of the probe. The first sub-module predicts the unsigned relative orientation between consecutive slices, while the second leverages a custom similarity model and a spatial context volume to determine the sign of this relative orientation. From the sub-modules predictions, slices orientations along the navigated trajectory can then be derived in real-time. Results demonstrate that registration error remains below 2.5 mm in 92% of cases over a 5-second trajectory, and 80% over a 25-second trajectory. These findings show that accurate, sensorless 2D/3D US registration given a spatial context is achievable with limited drift over extended navigation. This highlights the potential of AI-driven biopsy assistance to increase the accuracy of freehand biopsy.

This study introduces a novel approach to improve Cone Beam CT (CBCT) image quality by developing a synthetic CT (sCT) generation method using CycleGAN with a Vision Transformer (ViT) and an Adaptive Fourier Neural Operator (AFNO). &#xD;&#xD;Approach: A dataset of 20 prostate cancer patients who received stereotactic body radiation therapy (SBRT) was used, consisting of paired CBCT and planning CT (pCT) images. The dataset was preprocessed by registering pCTs to CBCTs using deformation registration techniques, such as B-spline, followed by resampling to uniform voxel sizes and normalization. The model architecture integrates a CycleGAN with bidirectional generators, where the UNet generator is enhanced with a ViT at the bottleneck. AFNO functions as the attention mechanism for the ViT, operating on the input data in the Fourier domain. AFNO's innovations handle varying resolutions, mesh invariance, and efficient long-range dependency capture.&#xD;&#xD;Main Results: Our model improved significantly in preserving anatomical details and capturing complex image dependencies. The AFNO mechanism processed global image information effectively, adapting to interpatient variations for accurate sCT generation. Evaluation metrics like Mean Absolute Error (MAE), Peak Signal to Noise Ratio (PSNR), Structural Similarity Index (SSIM), and Normalized Cross Correlation (NCC), demonstrated the superiority of our method. Specifically, the model achieved an MAE of 9.71, PSNR of 37.08 dB, SSIM of 0.97, and NCC of 0.99, confirming its efficacy. &#xD;&#xD;Significance: The integration of AFNO within the CycleGAN UNet framework addresses Cone Beam CT image quality limitations. The model generates synthetic CTs that allow adaptive treatment planning during SBRT, enabling adjustments to the dose based on tumor response, thus reducing radiotoxicity from increased doses. This method's ability to preserve both global and local anatomical features shows potential for improving tumor targeting, adaptive radiotherapy planning, and clinical decision-making.