Breast cancer, the most prevalent malignant tumor among women worldwide, exhibits substantial heterogeneity, which manifests as different molecular subtypes with different therapeutic and prognostic implications. Owing to existing studies focusing on either kinetic or spatial heterogeneity in isolation, this study proposed a Dual Heterogeneity Fusion Network (DHF-Net) that integrated both the kinetic and spatial heterogeneities from DCE-MRIs for diagnosing breast cancer molecular subtypes. Initially, a convex analysis of mixtures algorithm was employed to identify dynamic heterogeneity subregions by analyzing contrast enhancement patterns over time. Meanwhile, K-Means clustering was utilized for spatial analysis to delineate spatial heterogeneity subregions that reflected structural diversity within tumors. Then, the dynamic and spatial heterogeneity features obtained from a ResNet-based feature extractor were integrated using a dual-attention module that incorporated both cross- and self-attentions. Final molecular subtype diagnosis was performed by a Mixture of Experts (MoE) framework. Experimental results demonstrated the effectiveness of the DHF-Net on a publicly available TCIA dataset in two molecular subtype classification tasks.Clinical Relevance-This study preliminary exploits both the kinetic and spatial heterogeneity to predict breast cancer molecular subtypes, contributing to personalized treatment for breast cancer patients with different molecular subtypes.

Accurate diagnosis and classification of Central Nervous System (CNS) tumors, particularly meningiomas, pose significant challenges, especially when using DNA methylation profiling data. Limited sample availability and the high dimensionality of methylation data can limit robust analysis and the development of reliable algorithms. This study introduces a novel feature-based-driven augmentation strategy that effectively integrates the underlying probability density distribution, addressing class imbalance and distributional tail issues. We demonstrate the advanced utility of this method when applied to DNA methylation microarray datasets (raw IDAT files derived from archival meningioma tissue analyzed using the Illumina Infinium Methylation EPIC v2.0 BeadChip kit) for meningioma classification using a Wav2Vec2 transformer. Our data-aware approach outperforms a conventional copy-based benchmark augmentation technique, with significantly improved classification accuracies. Models trained on the red channel of methylation data, augmented by our method, achieved near-perfect accuracy of 98.75% (AUC=0.988), with the green channel achieving 75.3% accuracy, outperforming the benchmark augmentation method with 79.2% and 62.5% accuracies for the red and green signals, respectively. Notably, the enhanced performance in identifying intermediate meningiomas, an underrepresented class in our dataset, highlights the efficacy of our proposed augmentation technique.Clinical significance-Our proposed approach holds clinical significance by incorporating data-driven biologically informed augmentation to enhance CNS tumor classification, leveraging the full spectrum of DNA methylation diversity for more accurate tumor subtyping.

Magnetic resonance imaging takes an important part in breast cancer screening and has become the gold standard for clinical diagnosis. It has greatly facilitated the promotion of breast disease screening among a broad population of women with low cost and open space of low-field MR system. We proposed a new design of breast coil to improve the signal-to-noise ratio (SNR) of low-field breast MR images, which named a flared birdcage breast coil for its characteristics. Through deforming from the conventional cylindrical birdcage coil structure, our coil accommodates well human breast tissue while its inner horizontal radiofrequency field (B₁) is perfectly orthogonal to the vertical main field (B₀). The results of both electromagnetic field simulation and MR imaging experiments demonstrate excellent performance. Phantom imaging revealed significant SNR improvements of 50~60% in different planes compared to the commercial breast coil. Besides, we tried to load passive resonant rings to further improve the imaging quality and was confirmed.

Colon cancer is one of the deadliest types of cancer in the United States, with close to 50,000 projected deaths in 2024. The disease requires early diagnosis to optimize chances of survival by enabling timely administration of treatment. To investigate the key non-genetic (NG) factors influencing the onset of colon cancer and evaluate how genetic factors enhance the performance of machine learning (ML) models in predicting incidence, we incorporated polygenic risk scores (PRSs) alongside NG data in ML models to predict 10-year incident risk prediction of colon cancer using data from the UK Biobank. This approach enabled us to assess the added predictive value of PRSs in multi-modal models in estimating the 10-year risk of developing colon cancer over NG data alone. Moreover, our research focused on identifying the most relevant and predictive PRS and validating them using a robust ML framework. To ensure the robustness, we restricted the cohort to White British individuals to minimize ancestry-related heterogeneity. PRSs have proven effective in enhancing disease prediction for conditions such as breast cancer, myocardial infarction, and schizophrenia, reinforcing their relevance in clinical research. Exploring six PRSs, our goal was to minimize false negatives while simultaneously maximizing area under the receiver-operating characteristic curve (AUC), in order to improve early detection rates by identifying those who are at risk for colon cancer. This research shows that PRSs can be used to enhance overall predictive ability of ML models in colon cancer research over NG factors alone, bolstering the argument for incorporating PRSs into routine clinical practice. PRSs can also help minimize false negatives, a key feature for disease prediction models, as missed potential diagnoses are life-threatening.

This study presents an innovative approach based on the Robotic Ultrasound Diagnostic System (RUDS) for 3D reconstruction of organs and lesions from sequential 2D ultrasound slices. The RUDS comprises four main components: the Organ Tracking Robot (OTR) for multi-angle probe scanning, the Phantom Posture Robot (PPR) for optimizing probe contact with the abdomen, the Robotic Bed (RB), and the Robotic Supporting Arm (RSA). Leveraging the high flexibility and degrees of freedom in probe manipulation, the OTR controls probe rotation to capture and process continuous 2D slices, enabling precise 3D reconstruction of the kidney and associated lesions. By adjusting scanning speed and contact angles and analyzing spatial relationships within the 3D models, RUDS demonstrates superior accuracy in 3D modeling, supporting preoperative planning in High-Intensity Focused Ultrasound (HIFU) and minimizing risks to non-tumor tissue.Clinical Relevance- This enables precise 3D reconstruction of kidneys and lesions from 2D ultrasound images, supporting HIFU planning and enhancing ultrasound diagnostic accuracy.

Upper Aerodigestive Tract (UADT) cancers are prevalent, and often fatal, due to late diagnosis. Like many other biological tissues, UADT tissue consists of multiple layers, with cancer commonly developing in the epithelial layer, before spreading to deeper structures. Accurate identification of each layer can aid in early diagnosis, resulting in improved treatment outcomes. This study presents a bioimpedance sensing probe with optimized electrode placement for diagnosis of UADT cancers. The probe, designed to fit a laryngoscope's working channel, is equipped with four electrodes, allowing for a total of 12 impedance measurements. To determine the electrode position on the probe, a novel optimization method is used to increase the difference between sensitivity distributions for different measurements. Experiments performed using simulation validate the probe's effectiveness in retrieving more information to be used for tissue identification, demonstrating its potential for real-time, on-site cancer detection.

Accurate segmentation of tumors and cysts in dental panoramic radiographs is crucial for effective diagnosis and treatment. However, conventional CNN-based approaches face challenges due to indistinct boundaries, structural complexity, and noise from varying imaging conditions. To address these limitations, we propose Dentaldiff, a novel diffusion-based segmentation model. Our method introduces a dynamic feature fusion strategy and an iterative denoising mechanism to enhance global feature extraction and noise robustness. Additionally, modified DenseUNet was designed to improve segmentation performance. The model achieves state-of-the-art performance, with mean IoU of 0.61 ± 0.07 and Dice score of 0.75 ± 0.05, outperforming existing methods. Dentaldiff effectively handles complex anatomical structures and is the first known application of diffusion models to dental panoramic segmentation. Experimental results show that Dentaldiff achieves higher segmentation performance compared to CNN-based methods, particularly in challenging cases with indistinct boundaries and noise, suggesting its potential for broader clinical application.

This study evaluates the feasibility of Electrical Impedance Spectroscopy (EIS) as a minimally invasive technique to differentiate healthy lung tissue from peripheral pulmonary nodules and masses during Electromagnetic Navigation Bronchoscopy (ENB). The diagnostic capability of EIS is also assessed in distinguishing malignant from benign lesions. Significant differences (P < 0.001) were observed between healthy tissue and lesions in all impedance parameters (resistance, reactance, modulus, and phase angle). Differences between benign and malignant nodules and masses were observed in the complex plane plot, but some overlap between both occurred possibly due to similarities between adenocarcinomas and inflammatory nodules, primarily in terms of water content. The findings suggest that EIS could be implemented as a potential tool to improve biopsy guidance, reducing the need for additional X-ray imaging during the procedure. In addition, increasing sample size is needed to enhance diagnostic accuracy.Clinical RelevanceElectrical Impedance Spectroscopy (EIS) has proven to be an effective tool for distinguishing healthy lung tissue from peripheral lung nodules and masses. Using this technique for guiding the biopsy localization could lead to an improved diagnostic yield while reducing dependence on ionizing radiation-based imaging, such as X-rays.

Focused ultrasound (FUS) combined with microbubbles has emerged as an effective technique for enhancing drug delivery across the stringent blood-brain barrier (BBB). However, tuning up FUS-sonicating parameters that lead to the desired uniform drug distribution at the target spots remains a challenge due to underlying biophysiological complexity, particularly in multi-target scenarios. Here, we propose a novel system model that characterizes the relationship between external actuating signals and therapeutic responses within a Multiple-Input Multiple-Output (MIMO) framework. We observe the drug delivery pathway as a communication channel, where transvascular and intratumoral pharmacokinetics introduce effects equivalent to channel distortion and interference, further leading to non-uniform drug distribution and unintended accumulation in healthy tissues. To mitigate these effects, we develop an equivalent channel matrix and exploit a known concept of precoding scheme with the aim to optimize sonication parameters in a way to ensure uniform and effective drug delivery across all targeted regions. We numerically demonstrate our approach using available preclinical data on low-intensity FUS-mediated chemotherapy for brain tumors.

In addition to mammography and ultrasound, breast magnetic resonance imaging (MRI) is indicative in a number of cases. Both cost and low reader availability hinder the more common use of MRI. This paper builds on the previously published work on breast region segmentation in MRI and evaluates the possibility of automated detection of suspicious lesions in the breast region using radiomics breast tissue analysis. The processing pipeline assumes regular grid division of breast tissue and characterization of each image patch by radiomics features for further binary classification using Random Forest (RF) and XGBoost to differentiate between suspicious lesions and fibro-glandular tissue. The patch-wise results obtained reveal F1 scores ≥ 0.92, with balanced precision (0.94) and recall (0.95) for different patch sizes. Reassembling patch decisions supports positioning of the identified suspicious breast regions and patient-level decision making.Clinical relevance- With high incidence rates of breast cancer, an automated detection of suspicious lesions in MRI breast scans can offer a valuable support in patient prioritization and clinical decision making, especially in settings with low reader availability.