Breast cancer diagnosis has greatly benefited from the use of deep learning (DL) and machine learning (ML) methods on multimodal data, including mammography, magnetic resonance imaging (MRI), ultrasound, histopathology, genomic data, and clinical reports. Every modality presents its challenges and opportunities, demanding a thorough understanding of used methods for accurate and early diagnosis. Our review provides an exhaustive survey of the latest advances in DL and ML applications to detect, classify, and segment breast cancer in these modalities. Our work also addresses basic issues of data limitations, model generalizability, and interpretability. To overcome these issues we explore the state-of-the-art Vision-Language Models (VLMs) as an emerging approach for multimodal learning. Through incorporation of the current advances and identification of areas of future improvement, this review is a precious source of information to researchers and practitioners. The findings highlight the importance of interdisciplinary collaboration, application of standardized data sets, and continuous improvement of the methodology to facilitate clinical uptake of AI-based breast cancer diagnostic systems.

Background: In this study, we examined the effectiveness of transfer learning in improving automatic segmentation of brain metastases on magnetic resonance imaging scans, with potential applications in preventive exams and remote diagnostics. Methods: We trained three deep learning models on a public dataset from the ASNR-MICCAI Brain Metastasis Challenge 2024, fine-tuned them on a small private dataset, and compared their performance to models trained from scratch. Results: Results showed that models using transfer learning performed better than scratch-trained models, though the improvement was not statistically substantial. The custom Tversky and Binary Cross-Entropy loss function helped manage class imbalance and reduce false negatives, limiting missed tumor regions. Medical experts noted that, while fine-tuned models worked well with larger, well-defined tumors, they struggled with tiny, scattered tumors in complex cases. Conclusions: This study highlights the potential of transfer learning and tailored loss functions in medical imaging, while also pointing out the models’ limitations in detecting very small tumors in challenging cases.

The research was aimed at analyzing modern algorithms for diagnosing lesions in multiple sclerosis on MRI images. Multiple sclerosis is a severe disease of the central nervous system and ranks first among the causes of disability in patients of young working age. In connection with the development of computer vision and machine learning technologies, the relevance of using these technologies for medical diagnostics is growing. Such approaches are necessary for the effective development and implementation of diagnostic systems using artificial intelligence. Modern algorithms and models for lesion segmentation were selected and implemented. Material and methods. The paper presents CV features of diagnosing multiple sclerosis on MRI images, existing data sets: ISBI-2015, MSSEG-2016, MSSEG-2021; existing algorithms and models for lesion segmentation: U-Net, nnU-Net, TransUnet, TransBTS, UNETR, Swin UNETR. Results and discussion. The architectures and models of nnU-Net, UNETR, Swin UNETR were trained and compared at ISBI2015 with various parameters and loss functions. Four MRI sequences were used: T2-WI, T2-FLAIR, PD, MPRAGE. Lesion segmentation was approved by certified experienced neuroradiologists. Conclusions. The approaches described in the paper including data processing, model training, and results analysis, focused on the selection and development of high-quality computer vision algorithms for identifying multiple sclerosis lesions in MRI images. Identification and segmentation of demyelination foci is a necessary step for diagnosing the disease, as well as for calculating and interpreting more meaningful indicators of disease severity and progression.

Accurate segmentation of breast tumors in dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is essential for precise diagnosis, treatment planning, and quantitative analysis. While deep learning methods have achieved strong performance in controlled research settings, their ability to generalize across diverse clinical datasets remains underexplored and poses a major barrier to clinical adoption. In this study, we evaluate the cross-dataset generalizability of a 3D Residual U-Net model using the multicenter MAMA-MIA benchmark, which consolidates four publicly available breast MRI collections annotated by expert radiologists. A leave-one-out experimental design is employed, with three datasets used for training and validation, and the remaining dataset held-out for independent testing to simulate real-world deployment scenarios. Model performance is assessed using Dice coefficient, Precision, and Recall, alongside quantitative analysis of tumor volume estimation accuracy. The best Dice score achieved by our model was 0.683 when tested on the NACT subset. Results show a consistent degradation in segmentation accuracy when models are applied to unseen datasets, indicating that performance declines significantly outside the distribution of the training data. The most pronounced drop occurs when the DUKE dataset serves as the held-out test set, where the model struggles to adapt to differences in pre-release preprocessing strategies. A targeted qualitative review of 160 representative scans further reveals key factors contributing to both successful and failed segmentations, including variations in image field of view, temporal enhancement patterns, acquisition era, and artifact prevalence. Overall, these findings underscore the importance of accounting for dataset heterogeneity, domain shift, and standardized preprocessing in the development of robust, clinically deployable breast MRI segmentation models capable of generalizing across institutions and imaging protocols.

Brain metastasis is a severe and complicated cancer that requires timely screening and follow-up approaches using precise segmentation techniques for effective surgical intervention, radiation therapy, or monitoring the disease progression. In this study, we explore the efficacy of transfer learning in improving the automatic segmentation of brain metastases on Magnetic Resonance Imaging scans with the subsequent possibility of using it in clinical practice for preventive examinations and remote diagnostics. We train three deep learning models on a public dataset from the Brain Tumor Segmentation 2024 Challenge, then fine-tune these pretrained models on a small private dataset and compare them to models trained on private data from scratch. The results indicate that models utilizing transfer learning achieve superior accuracy and generalization in segmenting brain metastases. We also observed that the custom loss function significantly enhances performance compared to the default configuration. This study highlights the importance of leveraging transfer learning in medical imaging to address challenges associated with small, specialized datasets.