The transperineal approach in prostate interventions is increasingly favored for its lower infection risk and enhanced targeting accuracy. However, its precision can be affected by factors such as needle deflection, prostate deformation, and operator variability. Numerous studies have proposed models of needle deflection based on needle tip orientation. However, we hypothesize that anatomical location of the prostate cancer has a more significant influence on the direction of needle deflection than the needle tip orientation of clinically available needles. To test the hypothesis, we analyzed data from 231 in-vivo, transperineal, robot-assisted MRI-guided prostate biopsies using a mixed-effects logistic regression model to examine the association between needle tip orientation, the anatomical location of the prostate target and the resulting direction of needle deflection. The analysis shows that the needle tip orientation is not a significant predictor of needle deviation direction, while the anatomical location of the prostate targeted during each insertion appeared as a significant predictor of deflection direction (p < 0.001, OR = 10.89). We conclude that the anatomical location is a key determinant of needle deflection, while needle tip orientation has minimal impact. These results provide critical insights into the mechanics of needle-tissue interaction, potentially informing strategies to improve targeting accuracy in transperineal prostate interventions.

Accurate tumor margin identification is essential for ensuring complete removal and minimizing the risk of regrowth in cancer surgeries. As pathological changes in tissue results in altered electrical properties, impedance measurements can support surgeons to identify tumorous regions. In comparison to simple impedance sensors, multielectrode sensor arrays allow for a planar impedance mapping of the underlying tissue, enabling a precise tumor margin estimation. However, the measured impedance values do not take into account the different electrode configurations used for each measurement and result in inaccurate electrical maps. In this work, we first introduce an analytical form for the so-called geometry factor for general four-electrode measurement configurations, allowing for a scaled impedance (impedivity) mapping. Then, by simulating impedance measurements with a quadratic multielectrode sensor array, the corresponding impedivity map is created by scaling each impedance value for each convex electrode configuration with the associated geometry factor. We simulate different tumor scenarios to compare our novel impedivity maps to state-of-the-art impedance maps. The tumor margins are estimated using Otsu's method, based on the pixel-wise variations within the generated maps. The results demonstrate that the new impedivity map consistently outperforms the impedance map in tumor margin estimation, with accuracy improvements across all simulation scenarios. For surface tumors, the impedivity map achieves an accuracy of 90.3 % compared to 85.7 % for the impedance map. For deeper-seated tumors, the accuracy of the impedivity map remains consistently high, ranging between 84 % and 91 %. In contrast, the impedance map achieves a maximum accuracy of only 61 %, with its lowest performance dropping to 36 %. In a next step, these findings need to be validated experimentally.

Immunohistochemical (IHC) staining is a crucial technique for diagnosing and formulating treatment plans for breast cancer, particularly by evaluating the expression of biomarkers like human epidermal growth factor receptor-2. However, the high cost and complexity of IHC staining procedures have driven research toward generating IHC-stained images directly from more readily available Hematoxylin and Eosin-stained images using image-to-image (I2I) translation methods. In this work, we propose a novel approach that combines the predictive capabilities of three state-of-the-art I2I models to enhance the quality and reliability of synthetic IHC images. Specifically, we designed a Convolutional Neural Network that takes as input a four-dimensional input comprising the outputs of three distinct models (each contributing an IHC prediction, which is an RGB three-dimensional output for each) and produces a final consensus image through a fusion mechanism. This ensemble method leverages the strengths of each model, leading to more robust and accurate IHC image generation. Extensive experiments on the BCI dataset demonstrate that our approach outperforms existing single-model methods, achieving superior Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) metrics. All of our code is available at: https://github.com/arshamhaq/BCI-fusion.Clinical RelevanceImproving the quality of synthetic IHC images can potentially reduce costs and streamline the diagnostic process, ultimately benefiting patient outcomes.

Current commercial stainless-steel speculum based cervical examinations often cause discomfort, which may lower patient compliance for future screening. This study introduces a flexible silicone probe sheath inspired by menstrual cups as an alternative to the stainless-steel speculum. The silicon probe sheath expands within the vaginal canal to permit access to the transvaginal imaging probe while minimizing the deformation of the vaginal canal. A structural analysis was performed to tradeoff between the deformation and the factor of safety (FoS) of the probe sheath, which showed a thickness of 1.60 mm as the most optimal. The silicone probe sheath provides a patient-friendly alternative to the speculum, potentially reducing discomfort and increasing participation in cervical cancer screening. This design encourages self-insertion of the silicon probe sheet by the patients, thereby enhancing accessibility and patient compliance in clinical settings.

Nearly 90% of women called back after a suspicious mammogram do not have breast cancer, yet false positives lead to unnecessary biopsies, psychological distress, additional healthcare visits, and significant financial costs. The authors present an in-clinic compression device compatible with commercial ultrasound systems, utilizing Sub-Hertz Analysis of Visco-Elasticity (SAVE) to assess tissue viscoelasticity as a biomarker for malignancy. SAVE combines ultrasound imaging with a computer-controlled axial compression device to measure viscoelastic parameters, improving diagnostic accuracy and reducing subjectivity. The device integrates mechanical, electrical, and software components into a user-friendly tool, automating the SAVE method. Initial tests with breast phantoms showed strong agreement with a standard mechanical testing instrument, demonstrating high accuracy in key parameters such as the primary time constant, T₁ (mean absolute error: 5.78%). These results underscore the device's potential for accurate measurement of viscoelastic parameters of tissue with the goal of characterizing breast masses in vivo.Clinical Relevance- This novel compression device, facilitates the accurate application of SAVE method for in-clinic breast lesion assessment, aiming to improve specificity and reduces false positives. Its integration with standard ultrasound systems provides a cost-effective, non-invasive method to distinguish benign from malignant lesions, reducing unnecessary biopsies and patient anxiety. This advancement has the potential to enhance diagnostic workflows and patient care.

Skin cancer is one of the most common cancers worldwide and is primarily diagnosed through visual examination. With the availability of large amounts of dermoscopic data, recent advancements in artificial intelligence (AI) have achieved remarkable accuracy in skin cancer classification. However, due to the black-box nature of deep learning models, dermatologists often struggle to understand the underlying decision-making process, limiting the transparency and interpretability of AI-driven diagnoses. In this work, we investigate advancements in Prototypical Part Networks (ProtoPNet) to skin cancer detection by applying the Pixel-Grounded Prototypical Part Network (PIXPNET), designed to address the challenge of pixel-space mapping in prototype projection. The PIXPNET architecture was trained and evaluated to assess its generalizability. Our results show that PIXPNET significantly outperforms ProtoP-Net for skin cancer detection in a multi-class classification setting. Additionally, we analyze the learned prototypes to assess their relevance to input images, demonstrating improved interpretability compared to its counterpart, ProtoPNet.

Tiny Machine Learning (TinyML) offers a trans-formative approach to efficient, intelligent cancer diagnostics on edge devices. However, designing optimized TinyML models for breast cancer diagnosis using Microwave Imaging datasets remains a challenge due to domain-specific customization requirements. To address this, we propose a novel Knowledge Distillation framework, where a teacher model, built using residual convolutional neural networks, transfers its knowledge to a lightweight student model, enhancing efficiency while maintaining high accuracy. The teacher model achieves a 95.42% accuracy on test data. Through knowledge distillation, the student model attains 95.32% accuracy while achieving a 96% reduction in model size, significantly enhancing computational efficiency without compromising performance. Without distillation, the student model reaches a testing accuracy of 86.5%. A major contribution of this work is the generation of a high-quality synthetic breast cancer dataset to address the scarcity of microwave imaging data. We employ a Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) to synthesize realistic breast cancer scans, ensuring reliability through rigorous validation with a one-class Support Vector Machine. By training all models on a combination of real and synthetic datasets, our approach enhances robustness, making TinyML-powered breast cancer detection more viable for real-world deployment.Clinical RelevanceThis work enables efficient, accurate breast cancer detection on edge devices, supporting early diagnosis in clinical settings.

Neoadjuvant immunotherapy is an emerging treatment for lung cancer. However, its efficacy varies significantly due to the complexity of the immune microenvironment, and there is a lack of effective methods for predicting individualized efficacy of neoadjuvant immunotherapy. In this study, we propose a novel multi-instance learning model GLAM to mine fine-grained tertiary lymphoid structure (TLS) features and global immune microenvironment features from H&E-stained whole-slide images (WSI) to predict multiple clinical end-events that can reflect individualized short-term and long-term efficacy of neoadjuvant immunotherapy in lung cancer. We first train a network to predict TLS maturity, a prognostic indicator for immunotherapy, using a semi-supervised learning method. Then we combine fine-grained TLS features and global immune features via cross-attention and build a multi-instance learning model with self-attention to predict efficacy end-events. This study includes 194 lung cancer patients with post-operative WSI who received neoadjuvant immunotherapy, and the GLAM model demonstrates strong predictive performance across both short-term and long-term efficacy endpoints. For short-term efficacy endpoints, it achieves AUC=0.951 for predicting major pathological response, and AUC=0.864 for predicting pathological complete response. For long-term efficacy endpoints, it achieves AUC=0.911 for predicting 2.5-year recurrence status, and C-Index=0.805 for predicting individualized recurrence time.Clinical Relevance- This study provides a new method for predicting individualized short-term and long-term efficacy of neoadjuvant immunotherapy, which helps guiding personalized treatment planning for lung cancer patients undergoing neoadjuvant immunotherapy.

Magnetic resonance imaging provides detailed visualization of healthy and abnormal tissues, making it an essential tool for accurate diagnosis. Recent advancements in MRI Image super-resolution networks have shown promising potential. However, the effective aggregation of multi-scale textural details and high-frequency information, which is critical to achieving accurate reconstruction and subsequent clinical applications, remains a significant challenge. To address this limitation, we propose a Multi-scale Feature Aggregation Network (MFAN) for brain MRI image super-resolution. To ensure the selection of the most informative feature channels and spatially significant regions, the proposed network incorporates Channel and Spatial Attention (CSA) mechanisms for shallow feature extraction. In addition, we introduce a Multi-scale Feature Aggregation Attention Block (MFAAB), which extracts and fuses diverse features from multiple pathways, thereby enabling more accurate MRI reconstruction and enhancing the reliability of clinical diagnoses. Experimental results demonstrate that MFAN surpasses state-of-the-art methods on the BraTS 2018 and Brain Tumor datasets. Specifically, on the BraTS 2018 dataset, our model achieves PSNR improvements of 1.054 dB and 0.609 dB and SSIM gains of 0.0128 and 0.0059 at ×2 and ×4 magnifications, respectively.Clinical relevance- The proposed MFAN offers a substantial advancement in brain MRI image super-resolution, positioned to address critical challenges in clinical neuroimaging. Accurate reconstruction of high-resolution images is vital for the reliable detection and diagnosis. By effectively aggregating multiscale textural information and enhancing structural details, MFAN improves diagnostic precision while reducing reliance on repeated scans or high-field MRI systems.

The clinical monitoring of the central nervous system (CNS) tumors is challenging due to the limited availability of predictive imaging tools and the complexity of tumor progression. Meningiomas, which account for 35 % of CNS tumors, require precise growth assessment to allow precise clinical decision-making. Current manual assessment protocols rely on recurrent imaging using magnetic resonance imaging (MRI) and computed tomography (CT) to monitor tumor growth and the evolution of intrinsic features over time. However, accurately predicting tumor growth risks and rates has been a difficult challenge utilizing contemporaneous technology.In this study, we introduce a novel application of machine learning (ML) for predicting meningioma growth risks - categorizing tumors as growing, stable, or shrinking - and further estimating their volumetric growth rate using neurosurgeon-assessed clinical features derived from only one single imaging timepoint. We used 12 features including calcification, cerebrospinal fluid (CSF) plane, oedema, location, T2 intensity, regularity, sex, ethnicity, and age at diagnosis, derived from MRI and CT scans of 336 patients treated at Auckland City Hospital. Importantly, no volumetric data were included amongst the features to ensure non-biased model reliability. Our results demonstrate that machine learning can accurately predict meningioma growth risk and volumetric growth rates, achieving remarkably high accuracies exceeding 99 %. Among the tested ML models, k-nearest neighbors (KNN) consistently outperformed the others in both prediction tasks under 5-fold and 10-fold cross-validation schemes.Clinical relevance-This study establishes a paradigm for using ML to predict meningioma growth risk and progression, which provides an opportunity for improved patient-specific tumor monitoring and potentially for early intervention.