Daily Archives: July 20, 2026

Authors: Bhavana N, Kumar Siddamallappa. U, Neelamma. G, Anusha Jajur. J

Abstract: Kidney disease is a significant health concern worldwide, and identifying renal abnormalities at an early stage is essential for preventing disease progression and improving treatment outcomes. Manual interpretation of medical images can be time-consuming and may vary depending on clinical expertise. To address this challenge, this study presents a convolutional neural network (CNN)-based kidney disease classification system that incorporates transfer learning with ResNet101 and VGG16 architectures. The proposed model is trained and evaluated using an augmented dataset of renal ultrasound and CT images categorized into four classes: normal, cyst, stone, and tumor. Image preprocessing and data augmentation techniques are applied to improve image quality, increase dataset diversity, and enhance the model's generalization capability. By utilizing pre-trained deep learning models, the system effectively extracts meaningful image features while reducing training time and computational complexity. Experimental evaluation shows that ResNet101 achieves a classification accuracy of 97.4% while VGG16 achieves 95.8 %. Performance assessment using precision, recall, and F1-score further confirms the reliability of the proposed approach for multi-class kidney disease classification. The developed framework demonstrates the effectiveness of transfer learning for medical image analysis, particularly when labeled datasets are limited. In addition, the system has the potential to support radiologists by providing faster and more consistent diagnostic assistance, leading to improved clinical decision-making. Overall, the proposed approach ResNet101 offers an efficient and accurate solution for automated kidney disease detection and classification using deep learning techniques.

DOI: http://doi.org/10.5281/zenodo.21453407

Published by: vikaspatanker

Authors: Geofrey Nyabuto, Peters Anselemo Ikoha, Samuel Mungai Mbuguah

Abstract: Background: Binary depression screening does not distinguish patients with mild symptoms from those with clinically urgent symptom burden. Among people living with HIV (PLHIV), depressive-symptom severity may reflect nonlinear interactions among anxiety, immune status, treatment adherence, stigma, behavioural exposures, and demographic characteristics. Routine electronic medical records (EMRs) provide an opportunity to model these relationships using computationally reproducible methods. Objective: To develop and internally validate a polynomial regression machine learning model for predicting continuous PHQ-9 depressive-symptom severity among PLHIV using routine HIV-care EMR data. Methods: A cross-sectional patient-level analytical dataset was constructed from 54,301 de-identified EMR records from Bungoma and Busia counties, Kenya. Fifteen clinical, treatment, psychosocial, behavioural, and demographic predictors were processed using training-derived imputation, encoding, log transformation, and standardisation. Stratified training (n=38,010), validation (n=8,145), and held-out test (n=8,146) partitions were used. Ordinary least-squares regression with degree-1, degree-2, and degree-3 polynomial feature expansions was compared using R², adjusted R², mean absolute error (MAE), mean squared error (MSE), and root mean squared error (RMSE). Results: The degree-3 model comprised 816 fitted parameters and achieved the strongest held-out performance: R² 0.753, adjusted R² 0.730, MAE 1.808, MSE 6.359, and RMSE 2.522 PHQ-9 points. The linear model achieved R² 0.729, MAE 1.939, and RMSE 2.641. Mean and median cubic-model residuals were 0.040 and 0.005, respectively, although the maximum positive residual was 15.186, indicating important underprediction in a small number of high-severity cases. The largest reported terms were anxiety × CD4 (β=−0.444) and anxiety² (β=0.413). Conclusions: Degree-3 polynomial regression modestly improved prediction of PHQ-9 depressive-symptom severity over linear and quadratic alternatives. Its average error may support broad risk stratification, but threshold crossing, model complexity, coefficient instability, and severe case underprediction preclude autonomous clinical use.

DOI: http://doi.org/

Published by: vikaspatanker

Authors: Ambuj Kumar Misra

Abstract: Accurate crop yield prediction is fundamental to food security planning, resource optimization, and climate resilience policy. This study presents a comprehensive secondary data approach to predicting corn (Zea mays L.) yields across the contiguous United States by integrating multi-source datasets including National Oceanic and Atmospheric Administration (NOAA) climate records, the Soil Survey Geographic Database (SSURGO), USDA National Agricultural Statistics Service (USDA-NASS) historical yield data, and MODIS-derived Normalized Difference Vegetation Index (NDVI) values spanning 2000–2022. We evaluate and compare six predictive modeling frameworks—Linear Regression, Support Vector Regression (SVR), Random Forest (RF), Gradient Boosting (GB), Long Short-Term Memory (LSTM) neural networks, and a hybrid CNN-LSTM ensemble. The hybrid CNN-LSTM model achieved the highest predictive accuracy with an R² of 0.93 and a Root Mean Square Error (RMSE) of 5.4 bu/acre, substantially outperforming the baseline linear regression (R² = 0.61, RMSE = 18.4 bu/acre). Growing Degree Days, summer precipitation, and soil organic matter were identified as the three most influential predictors. Results demonstrate that rigorously curated secondary data, when combined with advanced machine learning architectures, can yield operationally reliable crop forecasts at county to regional scales without requiring expensive field campaigns. Implications for agricultural decision-making, early warning systems, and climate adaptation planning are discussed.

DOI: http://doi.org/10.5281/zenodo.21452245

Published by: vikaspatanker