
Computer-aided diagnosis systems for mammography require reproducible workflows that connect image preparation, model inference, operating-threshold selection, visual explanation and structured reporting. This study presents a health-technology workflow for mammography decision support based on a transparent and reproducible software prototype. The workflow was developed using the CBIS-DDSM public mammography dataset. It combines DICOM handling, photometric correction, CLAHE-based enhancement, ROI-based preprocessing, convolutional neural network classification, Grad-CAM visualization and automated PDF reporting. VGG16 and ResNet-50 were trained and compared using transfer learning, progressive fine-tuning, mixed-precision training and L1 pruning. Two operating modes were analysed: a sensitivity-oriented threshold and a discriminative threshold based on Youden’s index. On the independent CBIS-DDSM test set, VGG16 showed higher AUC than ResNet-50 (0.658 vs. 0.626) and was selected for deployment in the interactive prototype. At the sensitivity-oriented threshold, VGG16 reached 94.20
Cutaneous melanoma is the most lethal skin cancer, with incidence projected to rise by more than 50
To describe patient-reported media exposure, misinformation, advocacy awareness and support, and cancer-care policy priorities among patients with cancer in India, and to develop an exploratory patient-centred analytical framework incorporating formative composite measures of media influence and advocacy. A cross-sectional questionnaire-based survey of 1,189 unique patients with cancer was analysed. This includes demographic and clinical characteristics, sources and frequency of cancer-related information, perceived positive and negative effects of media, exposure to misleading information, advocacy awareness and support, and policy priorities. Composite indices included the Positive Media Score (PMS), Negative Media Score (NMS), Media Influence Index (MII), Advocacy Awareness Score (AAS), and Advocacy Support Score (ASS). Data analysis was performed using Python-based statistical coding. Clinician-led communication remained dominant, with doctors and hospital staff as the most common information sources. The Positive Media Score was 3.45 ± 0.70, the Negative Media Score was 2.68 ± 0.63, and the Media Influence Index was 0.77 ± 0.84, indicating higher positive-media composite scores than negative-media composite scores at the cohort level. Among participants reporting misinformation exposure, 45.7
To link reported healthcare cyber breaches to U.S. hospital records and examine whether affected counts increase in proportion to hospital size. HHS Office for Civil Rights (OCR) submissions from 2010–2025 were matched to annual CMS Provider of Services hospital records by name, state, and submission year. Models used high-scoring links to names classified as individual facilities, adjusted for hospital characteristics and submission era, and used facility-cluster bootstrap intervals. Outcomes were log (1+individuals affected) and log [(1+individuals affected)/(1+beds)] . Secondary analyses used Firth logistic regression. No manual linkage validation was performed. Of 566 hospital-like submissions, 354 met the acceptance threshold and 215 entered the primary cohort. The bed-count ratio was 1.39 (95
To evaluate the effect of streak artifacts resulting from arm-down positioning on single-photon emission computed tomography (SPECT) image properties when computed tomography attenuation correction (CTAC) is applied. The National Electrical Manufacturers Association International Electrotechnical Commission body phantom containing a cold sphere filled with 3:1 99mTc solution was scanned using SPECT/CT with and without an anthropomorphic arm phantom. CT scans for CTAC were performed at volume CT dose index (CTDIvol) settings of 0.6, 1.2, 2.5, and 5.0 mGy. SPECT images were reconstructed using 3D ordered subset expectation maximization. Image quality was evaluated using percent root-mean-square uniformity (