Extracting features from Whole Slide Images (WSIs) using pre-trained models is essential for computational pathology. However, this process usually requires substantial computational resources. Knowledge distillation aims to effectively transfer knowledge from complex, pre-trained teacher models to smaller, more efficient student models, which can be deployed in clinical scene with limited computational resources. Unfortunately, this process becomes especially challenging when teacher models come from different structures or modalities, leading to distributional gaps. To address these challenges, we propose a novel distillation method, named Efficient Computational Pathology Heterogeneous Model Knowledge Distillation (ECP-KD), which utilizes structure adapter layer and MIL adapter layer to bridge the distributional gap between teacher and student models. ECP-KD can effectively handle network mismatch problems with different structures in distillation and typical multi-instance learning tasks such as smart WSIs analysis tasks. We also incorporate cross-attention mechanisms for the fusion of multiple pre-trained models, enabling dynamic and scalable integration of various modalities. Experimental results on The Cancer Genome Atlas demonstrate that the proposed ECP-KD improves the performance of student models in survival prediction tasks, offering a speedup of up to 72x and reducing the number of parameters by up to 33x compared to larger ViT teacher models, with the state-of-the-art accuracy.Clinical relevance- This method enables more efficient and accurate in computational pathology tasks, making it applicable to resource-constrained clinical settings by improving model performance without compromising computational efficiency.

This study explores the application of various deep learning models for the segmentation of lung nodules using LIDC-IDRI. Unlike traditional approaches that utilize the full dataset, our work emphasizes the efficacy of training models on a filtered subset of 356 samples. Novel configurations, including attention mechanisms and advanced preprocessing strategies, were employed to optimize segmentation accuracy. Among the models evaluated, the DPLinkNet50 with a Channel Attention Bridge and ResNet backbone demonstrated the highest performance with a Dice score of 0.86, significantly outperforming conventional architectures. This work underscores the potential of leveraging data efficiency and tailored architectures in achieving robust segmentation performance, paving the way for improved computer-aided diagnosis in clinical settings.

Accurate classification of tumor immune microenvironment (TIME) subgroups is critical for predicting immunotherapy response and informing personalized treatment strategies. While immune subgroups are known to correlate with immunotherapy efficacy and prognosis, the underlying microenvironmental factors remain incompletely understood. In this study, we developed a machine learning-based classification model using FPKM-normalized RNA-Seq data from 440 immune-related genes. The model, trained with the eXtreme Gradient Boosting (XGBoost) algorithm on 7,300 samples, achieved a macro-balanced accuracy of 0.959 and a macro-balanced F1 score of 0.908 on an independent test set of 1,826 samples.Notably, the model also identified a seventh, predominant subgroup that exhibits mixed characteristics of the six established TIME subgroups, offering a new perspective on tumor heterogeneity. To support clinical and research use, the model has been deployed as a user-friendly web interface with integrated visualization tools, including Principal Component Analysis (PCA) and T-distributed Stochastic Neighbor Embedding (t-SNE), for classification and exploratory analysis. This tool has the potential to enhance immunotherapy research and facilitate more precise treatment planning.

Breast cancer diagnosis using ultrasound imaging remains challenging due to noise, variability in lesion appearance, and several artifacts. To address these concerns, this study proposes WAveHCT, a Wavelet-Attentive Hybrid Convolution-Transformer Network that uses wavelet decomposition and a hybrid architecture to diagnose breast cancers in ultrasound images. The proposed approach begins by applying anisotropic diffusion filtering to the ultrasound images, effectively reducing noise while preserving edge details. A ResNet50-based encoder backbone is then used to extract features from the wavelet-decomposed and anisotropic diffusion-filtered images. These features are integrated using a novel Wavelet-Attentive Feature Fusion (WAFF) module, enabling improved diagnostic performance. Further, the study introduces a hybrid block with convolutional and transformer layers. The transformer layers effectively capture global dependencies, while convolution operations preserve local feature representations. WAveHCT demonstrated superior accuracy, recall, F1-score, and AUC compared to existing methods. Heatmaps generated by WAveHCT exhibited improved localization of clinically relevant features, emphasizing its potential to assist radiologists in diagnosing breast cancers.Clinical relevance-Ultrasound imaging is a cost-effective, non-invasive and non-ionizing method of breast cancer screening. Therefore, developing advanced deeplearning-based tools for diagnosing breast cancer using ultrasound images can enhance radiologists' efficiency and reduce unnecessary invasive biopsies.

Multimodal medical image deformable registration is a critical foundation for liver tumor interventional therapy, assisting doctors in path planning and ablation efficacy evaluation. Research on this topic is still limited, especially regarding the challenges in liver multimodal deformable registration, such as significant image intensity differences and large organ deformations. In this paper, we propose DMCR--a novel multimodal multiscale image registration model based on a dual-stream Mamba with contrastive representation. During the encoding stage, images from different modalities are fed into distinct feature branches constructed using the Mamba architecture. In the registration stage, we propose a multiscale registration approach based on a decoding strategy, wherein the deformation field is progressively propagated and fused across multiple scales. Furthermore, we introduce a modality-invariant contrastive loss to guide the model in capturing intrinsic image features during the encoding stage, while disregarding modality-specific details, thus enhancing the effectiveness of subsequent registration. We trained and tested our model on multimodal liver ablation datasets from three different medical centers. The results demonstrate that our proposed model achieves superior registration performance compared to state-of-the-art registration methods and also exhibits better generalization capability.Clinical relevance- This model effectively performs multi-modal image deformable registration, assisting physicians in facilitating interventional procedures and evaluating multimodal therapeutic outcomes.

Cardiovascular toxicity constitutes a major adverse effect associated with cancer therapies. The optimal moment to evaluate and implement strategies for the prevention of cardiovascular toxicity in cancer patients is at the point of cancer diagnosis, preceding the commencement of oncological interventions. This study presents a machine learning pipeline for discovering pre-treatment prognostic biomarkers of asymptomatic cancer-therapy related cardiac dysfunction (CTRCD), using a dataset of 485 female patients diagnosed with breast cancer at the age of 55 or older. We examine a comprehensive set of clinical, biochemical, cardiac imaging markers and integrate them into a feature selection approach based on recursive feature elimination with cross-validation. The results identified a panel of significant features which exhibits an area under the ROC curve (ROC-AUC) equal to 0.75±0.07, including left ventricular ejection fraction, trastuzumab loading dose, body surface area, platelet count, left ventricular posterior wall diameter, white blood cell count, body mass index and interventricular septal diameter in ≥90% of the resampling iterations.Clinical Relevance- Early identification of high-risk patients for CTRCD enables proactive interventions, including the initiation of cardioprotective medications, adjustment of cancer treatments, and enhanced cardiac surveillance. Identifying biomarkers indicative of early cardiac injury allows clinicians to customize strategies aimed at preserving cardiac function and enhancing long-term outcomes for breast cancer patients.

Several studies indicate that deep learning models can learn to detect breast cancer from mammograms (X-ray images of the breasts). However, challenges with overfitting and poor generalisability prevent their routine use in the clinic. Models trained on data from one patient population may not perform well on another due to differences in their data domains, emerging due to variations in scanning technology or patient characteristics. Data augmentation techniques can be used to improve generalisability by expanding the diversity of feature representations in the training data by altering existing examples. Image-to-image translation models are one approach capable of imposing the characteristic feature representations (i.e. style) of images from one dataset onto another. However, evaluating model performance is non-trivial, particularly in the absence of ground truths (a common reality in medical imaging). Here, we describe some key aspects that should be considered when evaluating style transfer algorithms, highlighting the advantages and disadvantages of popular metrics, and important factors to be mindful of when implementing them in practice. We consider two types of generative models: a cycle-consistent generative adversarial network (CycleGAN) and a diffusion-based SynDiff model. We learn unpaired image-to-image translation across three mammography datasets. We highlight that undesirable aspects of model performance may determine the suitability of some metrics, and also provide some analysis indicating the extent to which various metrics assess unique aspects of model performance. We emphasise the need to use several metrics for a comprehensive assessment of model performance.Clinical relevance- Image-to-image translation models are used to augment the training sets of disease classifiers, which can supplement small datasets and potentially reduce bias. This can improve model generalisability, equitability and performance, and hence, patient outcomes. This work describes important factors that need to be considered to effectively evaluate the performance of image-to-image translation models in a trustworthy manner.

Gastric cancer (GC) is one of the most prevalent malignant cancers, with currently unsuccessful treatment strategies for patients. Increased PI3K/AKT/STAT-3 pathway activity has been observed in patients with GC. Morin (MRN), which is a flavonoid, exhibits significant anticancer activity by inhibiting the PI3K/AKT signaling pathway. However, monotherapy with MRN has faced challenges due to poor bioavailability and rapid elimination. This study investigated the combined effects of MRN and paclitaxel (PTX) on apoptosis induction and their molecular mechanisms in GC cells (HGT-1). After 24 h of MRN and PTX exposure, various assays were performed to assess the suppression of HGT-1 cell proliferation. These included cytotoxicity assessments, reactive oxygen species (ROS) level measurements, apoptotic morphological features, mitochondrial membrane potential (ΔΨm), nuclear fragmentation, and cell cycle analysis. Further, the effect of MRN and PTX on STAT-3 expression and various proliferation and apoptotic proteins was investigated using western blotting. The results revealed that the MRN and PTX combination significantly induced cytotoxicity, increased ROS levels, and altered ΔΨm, resulting in HGT-1 cell apoptosis (P<0.05). Furthermore, MRN and PTX treatment decreased the expression of oncogenic proteins, such as C-Fos, KRAS, and p-ERK1, in HGT-1 cells (P<0.05). The combination treatment inhibited PI3K, AKT, and STAT3 expressions, thereby suppressing proliferation and inducing proapoptotic protein expression in HGT-1 cells. Therefore, the combination of MRN and PTX could serve as a therapeutic approach for malignant GC treatment.

Pentobarbital (PB), a barbiturate anesthetic, is widely used in many clinical treatments, including management of seizures and preoperative sedation. However, its potential role in cancer therapy remains underexplored. The inhibitory effects of PB on breast cancer proliferation were investigated in this study. Breast cancer cell lines MDA-MB-231 and MCF-7 were used to evaluate the effects of PB on cell proliferation. Proliferating cell nuclear antigen (PCNA) overexpression and knockdown models were established to assess its role in PB-mediated proliferation inhibition. Cell proliferation was measured using methyl thiazolyl tetrazolium (MTT), colony formation, and bromodeoxyuridine (BrdU) incorporation assays. Protein and mRNA expression levels of PCNA and cell cycle-related genes were analyzed by Western blot and real-time RT-PCR, respectively. Statistical analysis was performed using Student's t-test or one-way ANOVA. PB was found to inhibit cell proliferation and regulate the expression of cell cycle-related genes compared to the control group. Further analysis revealed that PB downregulated the expression of PCNA. Overexpression of PCNA increased the proliferation of MDA-MB-231 cells, while PCNA knockdown suppressed it. Notably, overexpression of PCNA could partially restore the proliferative capacity of MDA-MB-231 cells that had been inhibited by PB. The findings indicate that PB, in addition to its established roles as a sedative and anesthetic agent, suppresses breast cancer proliferation through the downregulation of PCNA expression. These results provides new theoretical evidence supporting the potential application of PB in cancer treatment.

This study aimed to investigate the cytotoxic, morphological, and molecular effects of Tamoxifen (TMX) and Chlorambucil (CHL) on breast cancer (MDA-MB-231) and cervical cancer (HeLa) cell lines. The impact of these agents on metastatic behavior, apoptotic mechanisms, and gene expression profiles was examined in both two-dimensional (2D) and three-dimensional (3D) cell culture models.

Cells were treated with varying concentrations of TMX and CHL. Cytotoxicity was assessed using the XTT assay, and morphological changes were monitored by microscopy. Migration and invasion assays assessed metastatic potential. VEGFA expression was quantified by qRT-PCR. In 3D cultures, treatment responses were evaluated based on size reduction and structural changes in hydrogel-based spheroids. Docking analysis was conducted to determine binding affinities of TMX and CHL.

TMX and CHL exhibited dose-dependent effects on breast and cervical cancer cells. Combination treatment led to significantly greater reductions in cell viability compared to controls (p < 0.05). Moreover, VEGFA expression was markedly reduced in both 2D and 3D models (p < 0.05). These findings support the potential therapeutic value of TMX and CHL. Docking analysis revealed highly negative binding energies, consistent with in vitro results, indicating synergistic interaction at molecular and cellular levels.

TMX and CHL combination therapy demonstrated potent anti-cancer activity in breast and cervical cancer models, reducing cell viability, metastatic capacity, and VEGFA expression. These results suggest that TMX and CHL, when used together, may represent a promising strategy for developing synergistic and targeted cancer therapies. Further in vivo and clinical validation is warranted.