Biomedical Physics & Engineering Express

This study aims to assess women's perceptions of artificial intelligence (AI) used in breast cancer screening in France by examining their knowledge of AI and the barriers to their participation in organized screening. The results of a survey conducted in June 2025 among a national sample of 2000 women (aged 40-75) reveal limited participation and persistent concerns among women. Nevertheless, despite a low awareness of specific AI applications, a large majority of the women surveyed are very favorable to the use of AI in breast cancer diagnosis, even considering it a lever to increase screening participation.

With the emergence of electroencephalography (EEG) as a tool in the cognitive domain, new demands are being placed on the technology to keep up with functional applications, especially in the context of at-home neural monitoring. New use cases have fostered development of wearable EEG (wEEG) devices: portable, low-cost headsets used for EEG monitoring. This evolution of technology and application has not been accompanied by development in technology evaluation, often relying on function-agnostic markers to assess devices for efficacy in this new space. With current methods limited in scope, this study designed, tested and evaluated a novel functionally-focused comparative protocol for wEEG devices. Eight participants undertook a protocol for the evaluation of four established wEEG devices, assessing cognitive resolution and general usability. Compared to a well-established traditional analysis method (eyes open/eyes closed protocol), the novel design proposed here enabled the same analysis of headset resolution, while also providing additional context into user preferences and opening downstream possibilities for specific cognitive insights. Future research could enable the development of this protocol into a standardised method to ensure the performance of wEEG technology can satisfy emerging clinical needs.

Pediatric patients with cardiac implantable electronic devices (CIEDs) face limited MRI access due to RF-induced heating, and computational modeling is increasingly used to characterize this risk. The validity of these simulations, however, depends on pairing body models with clinically realistic lead configurations, guidance that is currently lacking. We retrospectively analyzed 302 CIED surgeries in 281 pediatric patients to derive weight-based constraints for simulation design. Weight alone discriminated epicardial from endocardial lead implantation with AUC = 0.90, and adding age and height yielded no improvement, supporting weight as a sufficient single-parameter selection metric. The probabilistic crossover between approaches occurred at 44~kg, substantially higher than the 10 to 15~kg threshold commonly cited in the literature, with a broad transition zone of 21 to 66~kg in which both lead types were routinely used. Lead length was likewise weight-constrained: only 25~cm leads were observed in patients below 6~kg, and leads of 45~cm or longer were uncommon below 50~kg. These findings yield a three-tier framework, with epicardial-only configurations below 21~kg, dual configurations within 21 to 66~kg, and weight-thresholded lead lengths throughout, enabling MRI safety simulations to focus on clinically realizable anatomy and device combinations.

Background: Independent end-to-end verification of adaptive radiotherapy on MR-Linac systems is limited by the lack of patient-specific phantoms able to reproduce imaging and dosimetric properties from CT and MRI scanners. We present a method for automated generation of 4D, patient-specific, multi-material 3D-printable phantoms for quality assurance of adaptive radiotherapy on a 1.5T MR-Linac. Methods: Patient images were automatically segmented using a pretrained deep learning model. The segmented structures were converted into high-resolution 3D meshes and assembled into printable phantoms. A dosimeter holder was inserted at user-defined anatomical locations, with orientation optimized to avoid traversal across heterogeneous tissue interfaces. Physiological motion was incorporated by generating phantoms from images at different timepoints and interpolating deformation fields to create continuous 4D models. Multi-material organs designed by mixing a set of six polymers at various proportions were used to reproduce tissue-specific imaging properties. The properties of material mixtures were evaluated in a clinical CT simulator and a 1.5T MR-Linac. Results: The proposed workflow enables automated generation of anatomically realistic phantoms with several types of embedded dosimeters. A discrete search method was designed for placement and immobilization of OSLD, film, and ion chamber dosimeters. Calibration curves for Hounsfield units were derived through variations in radiopaque material content, while MR signal intensity was modulated by gel and tissue matrix mixtures. Patient-derived abdominal phantoms were fabricated at multiple scales while replicating internal anatomical detail. Multi-dimensional phantom generation enabled continuous representation of motion states with consistent mesh topology across phases. Conclusions: We demonstrate an end-to-end workflow for automated generation of 4D patient-specific phantoms for MR-Linac quality assurance. The method combines realistic anatomy, embedded dosimetry, multimodal imaging properties, and physiological motion within a single fabrication framework. This approachmay enable an improved validation of adaptive radiotherapy workflows in MR-guided treatment devices.

Optimizing radiotherapy dose distributions remain a resource-intensive bottleneck. Existing AI-based dose prediction methods often have limited generalizability because they rely on small, heterogeneous datasets. We present nnDoseNetv2, an auto-configured, end-to-end framework for dose prediction across diverse disease sites (head and neck, prostate, breast, and lung), prescription levels (1.5-84 Gy), and treatment modalities (IMRT, VMAT, and 3D-CRT). By integrating machine-specific beam geometry with 3D structural information, the framework is designed to generalize across varied clinical scenarios. A single multi-site model was trained on 1,000 clinical plans. On sites seen during training, performance was comparable to specialized site-specific models. On unseen sites (liver and whole brain), the model outperformed site-specific models, with mean absolute errors of 2.46% and 6.97% of prescription, respectively. These results suggest that geometric awareness can bridge disparate anatomical domains while eliminating the need for site-specific model maintenance, providing a scalable and high-fidelity approach for personalized radiotherapy planning.

Point-of-care ultrasound has transformed bedside diagnostics, yet current systems remain limited by rigid form factors, bulky external electronics and the need for skilled operators. Here we report a conformable ultrasound imaging patch that integrates a 1024-channel CMOS ultrasound application-specific integrated circuit (ASIC) directly beneath a conformable piezocomposite transducer array. The 10 mm X 8 mm, 1024-element ASIC contains on-chip transmit and receive beamforming, reducing the effective off-chip channel count by 16X while preserving image fidelity. Fabricated on a flexible polyimide substrate and bonded using anisotropic conductive film, the patch operates untethered from conventional ultrasound consoles and requires only a laptop for control and data acquisition. The device supports focused, plane-wave and diverging-wave transmission with steering over {+/-}30{degrees} in azimuth and {+/-}15{degrees} in elevation, achieving peak-to-peak acoustic pressures up to 7 MPa at a 4.4-MHz center frequency (mechanical index of 1.7), within diagnostic safety limits. Phantom experiments demonstrate three-dimensional imaging with axial and lateral resolutions (in both XZ and YZ planes) of 0.5 mm and 2 mm, respectively, and accurate contrast reproduction in tissue-mimicking phantoms. Human studies further demonstrate three-dimensional (3D) visualization of the internal jugular vein and carotid artery, as well as rib-shadow-free imaging of pleural motion during respiration. This work establishes a scalable architecture for chronic, wearable ultrasound imaging and highlights the potential of CMOS-integrated, conformable ultrasound systems for continuous physiological monitoring and remote diagnostics.

Background: Understanding how transcranial direct current stimulation (tDCS) affects brain activity critically benefits from the use of functional magnetic resonance imaging (fMRI) to measure the related BOLD (blood-oxygenation-level-dependent) signal changes. However, the small magnetic fields induced by the stimulation currents can cause artifacts in the fMRI images that can compromise findings from concurrent tDCS-fMRI studies. Objective: To identify how the current-induced magnetic fields affect fMRI data and establish a quantitative framework for evaluating their impact on concurrent tDCS-fMRI measurements. Methods: Magnetic fields induced by currents inside the head and electrode cables were calculated for a standard motor cortex montage. Their effects on echo-planar images (EPI) were simulated based on a framework derived from MR physics first principles and validated using phantom experiments. The framework was applied to artificially induce artifacts related to the tDCS current flow in current-free fMRI time series from 5 participants. These were compared to active runs from the same participants where tDCS intensity was varied in a block design. Results: Currents in the electrode cables were the main contributors to the current flow-related artifacts in the EPI images, which occurred both locally by causing geometric distortions and remotely by affecting the dynamic update of the scanner demodulation frequency. The artificially induced fMRI activations corresponded well to those measured during real tDCS on the single-subject level for intensities of 2 mA and higher. Conclusion: The current-induced magnetic fields can cause intensity changes comparable to typical BOLD responses. Their impact on the statistical results depends on the chosen experimental design (electrode locations, cable paths, imaging parameters, fMRI paradigm). The simulation framework provides a principled approach to evaluate the impact of these artifacts during the design and data analyses of concurrent tDCS-fMRI studies.

Visual prostheses face a critical miniaturisation challenge: converting photoreceptor signals to biologically appropriate retinal ganglion cell (RGC) stimulation patterns within the spatial constraints of intraocular implants. Existing systems rely on external microcontrollers for signal processing, limiting scalability for high-density pixel arrays. This paper presents an integrated per-pixel circuit architecture that directly converts photocurrent into frequency-modulated current pulses that match RGC activation thresholds. The design targets are established through NEURON computational modelling of red-green colour-opponent midget RGCs, identifying stimulation thresholds of +0.1nA to +3.5nA for depolarisation and -0.1nA for repolarisation. The proposed circuit combines a transimpedance amplifier, a voltage-controlled oscillator with a Schmitt trigger, and a current-controlled output stage to generate biphasic pulses within these thresholds. A complementary output provides lateral inhibition, reducing crosstalk between adjacent RGC stimulation sites. Photoreceptor integration is achieved using P3HT:PCBM organic photodiodes for cone-associated RGCs and phototransistors for rod-associated RGCs, validated through OghmaNano finite element simulations. The photodiode circuit produces output frequencies of 2.5Hz (dark) to 600Hz (100 W/m2), matching reported RGC response ranges. This architecture eliminates external processing requirements, enabling scalable high-density retinal prostheses design.

BackgroundRadiation protection is a fundamental component of diagnostic radiology to minimize occupational and patient exposure to ionizing radiation. ObjectiveTo assess compliance with radiation protection standards and identify key deficiencies in conventional radiology units. MethodsA descriptive cross-sectional study was conducted in the radiology department of Ibn Tofail University Hospital, Marrakech. Twenty-five occupationally exposed radiology personnel were included using exhaustive sampling. Data were collected using structured observation grids and self-administered questionnaires to assess compliance with radiation protection standards. ResultsThe study revealed major deficiencies, including absence of individual dosimetry, inadequate structural shielding, poor quality control procedures, and non-compliance with occupational safety standards. ConclusionRadiation protection practices remain suboptimal, requiring urgent institutional reinforcement of regulatory compliance and enforcement, monitoring systems, and training programs.

Computer vision and deep learning techniques, including convolutional neural networks (CNNs) and transformers, have increased the performance of medical image classification systems. However, training deep learning models using medical images is a challenging task that necessitates a substantial amount of annotated data. In this paper, we implement data augmentation strategies to tackle dataset imbalance in the VinDr-SpineXR dataset, which has a lower number of spine abnormality X-ray images compared to normal spine X-ray images. Geometric transformations and synthetic image generation using Generative Adversarial Networks are explored and applied to the abnormal classes of the dataset, and classifier performance is validated using VGG-16 and InceptionNet to identify the most effective augmentation technique. Additionally, we introduce a hybrid augmentation technique that addresses class imbalance, reduces computational overhead relative to a GAN-only approach, and achieves [~]99% validation accuracy with both classifiers across all three case studies.

PurposeTo present the design and validation of a lowcost, microcontrollerbased gas delivery system that automates fixed inspired respiratory stimuli for MRI experiments. MethodsThe system uses three solenoid valves controlled by an Arduinobased circuit to switch between premixed medical gas cylinders according to predefined timing protocols. By using the MRI scanner external timing signal, gas delivery can be synchronised with image acquisition. Both a permanently installed configuration and a portable enclosure were constructed using commercially available components, with a total material cost of approximately {pound}650. The system was integrated with a singleuse breathing circuit and evaluated using hypercapnic and hyperoxic stimulus paradigms. Endtidal oxygen and carbon dioxide were measured using a respiratory gas analyser and physiological responses were assessed using BOLD MRI at 3 T. ResultsThe system delivered reliable, repeatable gas transitions during MRItriggered protocols. During hypercapnia (n{square}={square}15), the mean increase in endtidal carbon dioxide was 8.7{square}{+/-}{square}1.8{square}mmHg from a baseline of 32.2{square}{+/-}{square}3.1{square}mmHg, producing a mean grey matter BOLD signal increase of 3.2{+/-}1.7%. During hyperoxia (n{square}={square}15), the mean increase in endtidal oxygen was 292.3{square}{+/-}{square}59.0{square}mmHg from a baseline of 114.5{square}{+/-}{square}10.7{square}mmHg, with an associated BOLD signal change of 1.2{+/-}1.7%. Across both protocols respiratory and BOLD responses were consistent across participants. ConclusionThis microcontrollerbased system provides an inexpensive and reliable method for administering fixed inspired respiratory stimuli with automated MRI synchronisation. It offers an intermediate option between simple manual systems and higher cost commercial gas blenders, making it well suited for technical and methodological studies in cerebrovascular reactivity, hyperoxiaBOLD and related applications.

Accurate segmentation of thin-wall organs-at-risk (OARs)-the cochlea, vestibular semicircular canals, internal auditory canal, tympanic cavity, and middle ear-is clinically relevant for head-and-neck radiotherapy planning, yet these small, thin-wall structures remain among the most challenging targets for automated delineation. Dual-frequency feature fusion is a promising direction for boundary-sensitive representation, but under the investigated FP16 FFT-FcaNet setting, we observe an approximately 863-fold activation-scale mismatch between the FFT and FcaNet branches, causing a nominal 5 percent residual coefficient to behave as an approximately 43-fold dominant term. We propose FreqFuseNet, which resolves this mismatch by normalizing the FcaNet branch to the FFT activation scale before residual injection with a fixed low-amplitude coefficient (beta = 0.05), restoring beta as an interpretable 5 percent residual-amplitude coefficient relative to the FFT feature scale. Under a controlled binary per-OAR ROI protocol on the SegRap2023 head-and-neck CT benchmark across 10 clinically prioritized thin-wall OARs, FreqFuseNet achieves Dice of 0.849, HD95 of 0.824 mm, and SDice@1mm of 0.959 in the primary seed, with comparable performance in an independent second seed (Dice 0.843, HD95 0.823 mm). FreqFuseNet yields statistically significant case-level aggregate improvements over 3D U-Net and MedNeXt-S (Wilcoxon p < 0.01 and p < 0.05, respectively), using only 29.7 million parameters versus 414.6 million for the full wavelet baseline.

Rationale and ObjectivesPublished estimates of benign breast disease (BBD) are derived mainly from clinical, surgical, screening-recall, or reduction-mammoplasty series. Forensic autopsy cohorts can reduce referral and symptom-selection bias, although they are not necessarily representative of the whole living population. We describe imaging-detected benign breast findings in the Sisyphus forensic autopsy cohort. Materials and MethodsConsecutive medico-legal autopsies of individuals aged 40 years or older were prospectively evaluated over a multi-year period at a medico-legal autopsy service in Portugal. Bilateral breast specimens obtained by subcutaneous modified radical mastectomy were examined with specimen digital mammography and ultrasonography. Findings were classified according to BI-RADS terminology. Lesions requiring tissue diagnosis in the post-mortem protocol underwent wire-guided or direct excisional biopsy. Female cadavers were analysed as the primary cohort; male cadavers were analysed separately as an exploratory subgroup. Proportions are reported with exact 95% confidence intervals (CIs). ResultsThe cohort included 291 cadavers: 217 women and 74 men. Among female breast specimens, 236/434 were BI-RADS 1 (54.4%; 95% CI, 49.6-59.1), 189/434 were BI-RADS 2 (43.5%; 95% CI, 38.8-48.4), and 8/434 were protocol-sampled suspicious findings (1.8%; 95% CI, 0.8-3.6). At the cadaver level, 99/217 women had at least one benign imaging finding (45.6%; 95% CI, 38.9-52.5). Mammographic benign findings were present in 91/217 women (41.9%; 95% CI, 35.3-48.8), dominated by calcifications; ultrasonographic benign findings were present in 51/217 (23.5%; 95% CI, 18.0-29.7), most often simple cysts and duct ectasia. Plasma cell mastitis-pattern calcifications were observed in 8/217 women (3.7%; 95% CI, 1.6-7.1). Male benign findings were less frequent (9/74, 12.2%; 95% CI, 5.7-21.8) and were dominated by benign lymph-node variants. All nine protocol-sampled lesions were benign at histology. Clinical breast examination identified 5/8 protocol-sampled female lesions (62.5%; 95% CI, 24.5-91.5). ConclusionIn this forensic autopsy cohort unselected for breast symptoms, benign imaging findings were common in women aged 40 years or older and less frequent in men. The results provide descriptive post-mortem imaging reference data, but lesion-specific estimates, especially rare entities, should be interpreted with caution because of small numerators, the older age profile, limited clinical history, and the original cancer-focused design of the Sisyphus study.

Purpose: To develop and evaluate a deep learning model for automated quantification of breast arterial calcification (BAC) on screening mammography and to assess whether AI-derived BAC burden predicts major adverse cardiovascular events (MACE) in women. Methods: In this retrospective study, 202,006 women who underwent screening mammography without history of MACE were included. A BAC segmentation model was trained on an expert-annotated dataset using a multi-task U-Net with a ResNet-18 encoder to detect and segment BAC. BAC burden was quantified as area (mm{superscript 2}) from model-generated masks using DICOM pixel spacing and categorized by tertiles into low, intermediate, and high. The PREVENT score and incident MACE were identified from electronic health records. Cox proportional hazards models were developed to evaluate AI-derived BAC burden and PREVENT score alone, and combined models for 5 - and 10-year cardiovascular risk prediction. Results: Among 202,006 women (mean age 54.8{+/-}11.7 years), 23.1% had AI-detected BAC, and 7,701 (3.8%) developed incident MACE during a median follow - up of 7.5 years. On the geographically held-out test set, the BAC model achieved an AUROC of 0.97, Dice score of 0.6678, and Pearson correlation of 0.961 between AI-derived and manually annotated BAC burden. BAC burden increased with age and was higher among women who developed MACE. Five - year MACE incidence increased across BAC categories from 1.5% in women without BAC to 6.9% in those with high BAC burden. BAC burden alone showed modest prediction of MACE, with 5-year and 10-year AUROCs of 0.661 and 0.650, respectively, while PREVENT achieved AUROCs of 0.781 and 0.771. Adding BAC to PREVENT produced minimal improvement in discrimination. Conclusion: Deep learning-based BAC quantification from routine mammography is feasible, accurate, and associated with future cardiovascular risk. Although BAC added little to PREVENT for overall discrimination, it may serve as a scalable opportunistic imaging biomarker to identify women at elevated cardiovascular risk and support preventive care.

Vagus nerve stimulation (VNS) is an established clinical therapy for drug-resistant epilepsy and shows potential for treating other conditions, including depression, rheumatoid arthritis, diabetes, and heart failure. However, stimulation often produces unwanted side effects such as hoarseness, coughing, and paraesthesia. In some cases, these effects limit the delivery of therapeutic stimulation levels and hinder the development of new neuromodulation therapies. Selective VNS (sVNS) offers a strategy to reduce off-target organ activation. MethodsThis work presents an NFC-controlled, wirelessly powered, battery-free, temporary implantable multichannel stimulation device, made with off-the-shelf components, enabling selective stimulation of specific regions of the vagus nerve (VN). The encapsulated device is suitable for short-term implantation in animals. Main resultThe sVNS device was investigated in a porcine (n = 4) trial and an n = 1 pilot human experiment. Selective bradycardia of 23.28 {+/-} 12.91% was achieved in pigs and 7.5% in the human participant. In the human, a clear separation of cardiac efferent and afferent stimulation was observed, with additional selectivity in laryngeal activity. Physiological effects of laryngeal and cardiac fibre separation were measured to be 231{degrees}. SignificanceGeometrically selective stimulation of VN fascicles has the potential to improve clinical outcomes, enhance therapeutic efficacy, and reduce stimulation-related side effects. This strategy may enhance neuromodulation approaches for the treatment of heart failure using VNS.

PurposeTo demonstrate the feasibility of prostate MRI at low-field and provide an optimised low-field prostate MRI protocol. MethodsWe acquired bi-parametric prostate MRI in 15 healthy male volunteers (mean age 62 years, 95% CI 57-67.5), including tri-planar T2-weighted imaging (WI) and axial diffusion weighted imaging (DWI), with parameters adhering to PI-RADS guidelines. Deep learning reconstruction (DLR) was used to enhance image quality. SNR and CNR were measured in the prostate peripheral zone (PZ) and transitional zone (TZ). SNR was compared between T2-WI with and without DLR, between DWI data acquired with different b-values, and between different calculated b-value images. Radiologists assigned PI-QUAL qualitative image scores. ResultsThe acquisition time (minutes:seconds) was 4:02 for axial T2-WI and 6:32 for DWI. Axial T2-WI had a median (IQR) SNR of 8.72 (7.27-9.67) in the TZ and 12.84 (12.00-15.87) in the PZ. CNR between the PZ and TZ was 5.84 (4.06-6.82). DLR substantially improved T2-WI quality (p<0.05). The optimised DWI protocol had b-values of 50 and 800 s mm-2, with a calculated b-value image at 1500 s mm-2. Median apparent diffusion coefficient (ADC) values were 1.70 and 1.44x10-3 mm2 s-1 in the PZ and TZ, respectively. All participants had at least acceptable diagnostic quality (PI-QUAL score[&ge;]2/3), of which ten (67%) had optimal diagnostic quality (PI-QUAL score=3/3). ConclusionProstate MRI is feasible at low-field, providing clinically acceptable acquisition times and diagnostic quality images.

Computed tomography is the largest contributor to population radiation dose from medical imaging, yet no diagnostic reference levels (DRLs) have been published from Kosovo or the Western Balkans. This retrospective audit analyzed all CT examinations performed on a 128- slice scanner at the University Clinical Centre of Kosovo between January and March 2026. After exclusions, 1,535 acquisitions from 1,092 patients across nine examination categories were analyzed. Local DRLs were defined as the 75th percentile and compared against German (BfS 2022) and Turkish (Kahraman et al., 2024) reference values. Head CT (n = 590) demonstrated CTDIvol 4.7% below the BfS DRL yet scan length 98.5% above the orientation value (median 25.8 vs 13 cm). Abdomen-pelvis CTDIvol matched the BfS reference while scan length exceeded it by 28%. Coronary CTA showed CTDIvol +377%, consistent with retrospective ECG gating. Excess scan length, not CTDIvol, is the major driver of elevated dose at this institution. The identified excesses are correctable through technologist landmarking training, protocol review, and enabling iterative reconstruction.

Abstract Background Immune-related adverse events (irAEs) involving the breast remain rarely reported. Purpose To characterize clinical and imaging features of camrelizumab-associated breast lesions (CABLs). Materials and Methods This retrospective dual cohort study (October 2019 to February 2026) included 196 female patients. Cohort A comprised 180 non-breast cancer patients; Cohort B comprised 16 breast cancer patients receiving neoadjuvant camrelizumab. Baseline characteristics, treatment response, and CT/MRI features were compared between CABL-positive and CABL-negative groups using Mann-Whitney U and chi-square tests. Results CABLs developed in 34.4% (62/180) of Cohort A and 93.8% (15/16) of Cohort B. CABL-positive patients were younger (median 50.5 vs 54.5 years; P = 0.006) and more often premenopausal (46.8% vs 26.3%; P = 0.009). The objective response rate was relatively high among patients with positive lesions; in Group A, the disease progression rate was lower in the CABL-positive group than in the CABL-negative group (3.2% vs 17.8%), whilst in Group B, the pathological complete response rate was as high as 53.3% (8/15). On CT/MRI, CABLs were predominantly multiple (62.5%), with well-defined margins and unrestricted diffusion. The predominant time-intensity curve (TIC) pattern was washout (46.7%). Median time to onset was 2-3 cycles (the second MRI scan); most lesions disappeared (40.3%) and shrank (46.8%) during follow-up. ADC values of lesions were significantly higher than those of primary tumors (1.847+/-0.284 vs 0.976+/-0.055 x10[-3] mm[2]/s; P < 0.001). Histopathology of four lesions revealed lymphocytic infiltration and fibrosis without malignancy. Conclusion CABLs are benign reactive changes driven by multiple factors. Their recognition prevents misinterpretation as disease progression, thereby avoiding unnecessary treatment discontinuation or biopsy.

Recent NCCN guideline highlighted AI-based mammographic risk prediction, but AI-based breast cancer detection remains questionable to translation. One barrier is current models often do not match routine clinical reasoning, which may add decision burden than benefits. In practice, radiologists compare current and prior mammograms while assessing breast density, bilateral symmetry, and lesion laterality. To align AI with this reasoning, we developed MuSTAF, a multi-task spatiotemporal attention fusion model for patient-level breast cancer classification from longitudinal full-field digital mammography. MuSTAF uses up to three recent mammograms, integrates temporal and cross-view information, refines suspicious-region features, and jointly predicts cancer status, breast density, and bilateral symmetry, with a separate laterality classifier for cancer-positive cases. In an internal case-control cohort (n = 351), MuSTAF achieved a cancer classification (AUC=0.84) exceeding all architecture-level baselines and published mammography AI models adapted to the same task (AUC [&le;] 0.81). Simultaneously, it achieved AUCs of 0.83/0.80 for density/laterality assessments, and removing these auxiliary tasks reduced cancer detection performance. On the external CSAW-CC dataset (n = 8,723), model performance improved from 0.72 to 0.88 when restricting cancer cases to those with latest exams within 60 days before diagnosis, showing that temporally distant labels may shift detection evaluation toward risk prediction. Longitudinal analysis further showed that three recent exams outperformed five exams internally (AUC = 0.84 vs 0.80) and externally (0.72 vs 0.66), indicating recent imaging evidence mattered more than remote history. Overall, MuSTAF model improved longitudinal mammographic cancer classification while providing auxiliary outputs, and clarified temporal factors for applying AI to screening detection.

International guidelines for low-frequency electromagnetic field exposure (LF EMF) are primarily intended to prevent substantiated adverse effects. In the frameworks, limits on internal electric fields are linked to external exposure levels through computational dosimetry. However, the relationship between internal electric fields and these adverse effects remains incompletely understood. In particular, current approaches often overlook the morphological complexity and diversity of cortical neurons, which may limit the realism of neuronal activation estimates used to support these assessments. This study evaluates LF EMF-induced neural activation using 25 morphologically realistic neuron models spanning all cortical layers, embedded within 11 detailed human head models. The internal electric fields were simulated for uniform magnetic field exposures (100 Hz-100 kHz) along the three anatomical directions, and excitation thresholds were computed using a multi-scale framework combining voxel-based dosimetry with biophysical neuron simulations. A real-world exposure scenario involving a child near an acousto-magnetic article-surveillance deactivator was also analyzed. Thresholds varied across cell type, morphology, cortical location, subject anatomy, frequency, and exposure direction, with L2/3 pyramidal, L4 basket, and L5 thick-tufted pyramidal cells showing the lowest thresholds. Despite this variability, all simulated thresholds were conservative with respect to the basic restrictions and dosimetric reference limits set by IEEE ICES and ICNIRP. The smallest margin occurred at 100 kHz, where the threshold remained a factor of 2.8 above the corresponding limit. These findings indicate that current LF EMF exposure limits remain conservative when evaluated using highly detailed, morphology-based CNS activation models.