Pulmonary nodule detection on three-dimensional computed tomography (3D-CT) imaging is critical for early lung cancer diagnosis in environments with limited radiologist resources. In this work, we evaluate two state-of-the-art pretrained detection models (LUNA16 and DLCSD) on a dataset of 324 anonymized full-chest CT scans acquired in Vietnamese hospitals, each accompanied by a free-text radiology report. We employ a keyword-based automated labeling pipeline, carefully handling negation and variable terminology, to generate pseudo-ground-truth for training and evaluation in a low-resource setting. Model performance is quantified at detection-level using multiple confidence thresholds (0.5, 0.8, 0.9, 0.95). Our results show high sensitivity (up to 96.7
The rapid growth of the fashion industry has intensified concerns about environmental degradation, resource consumption, and textile waste. Amid these challenges, rental clothing has emerged as a sustainable alternative that offers both ecological and economic advantages. However, the factors influencing consumers’ willingness to adopt rental fashion in emerging markets remain insufficiently understood. This study investigates the key drivers shaping consumer preference toward rental clothing, focusing on environmental concern, sustainability benefits, economic considerations, and hedonic motivations. A structured questionnaire was administered to 409 respondents across India, and the collected data were analyzed using reliability testing, principal component analysis (PCA), ANOVA, and multiple regression techniques. The findings reveal that sustainability and economic benefits significantly influence adoption intentions, while hedonic motivations also play a meaningful role. Environmental concern, although positively associated, emerged as a comparatively moderate predictor. The proposed model explains a substantial proportion of variance in consumer preference, highlighting the growing potential of rental fashion as a viable circular-economy model. By contextualizing behavioural factors within a technology-enabled retail environment, this research offers practical insights for digital rental platforms, sustainable fashion enterprises, and policymakers aiming to promote responsible consumption. The study also underscores the need for stronger awareness initiatives and improved platform trust to accelerate adoption in the Indian fashion market.
The identical criminal evidence that encounters unfavorable conditions in crime scenes is important for identifying fingerprints and supporting criminal investigation. Since crime scenes are uncontrolled environment, fingerprints are often compromised by factors such as dirt, moisture and physical damage. Traditional fingerprint analysis methods face challenges with noise, deformation and partial patterns, which limit their reliability in low evidence processing. To address these boundaries, this study benefits from deep learning and advanced imaging techniques to increase the recovery and classification of criminal evidence. The proposed deep learning method integrates the construction correlation and is evaluated on several data sets, including a simulated fingerprint for crime. The results show that the recovered and located fingerprints get better accuracy than data set adjustment or unpleasant data sets. Experimental analysis reveals significant improvement in image quality, an average PSNR increase of 2.3 db for SD -302A, 8.6 db for SD -302M and 0.4 db for SD -302P. In addition, SNR reforms of 1.7 db, 3.7 db and 0.3 db were observed, while the MSE values for the restored fingerprint images were reduced by 0.8, 1.4 and 0.2 respectively. These advances highlight deep learning's ability to remove the challenges of analyzing the fingers of the offended criminal evidence, offering stronger forensic checking equipment.
The need for effective and efficient detection of weeds during rice cultivation is becoming more imperative for sustainable agricultural practices. Other weeds that seriously affect the development of rice are Barnyard Grass, Jungle Rice, Cockspur Grass, Chinese Sprangletop, and Crowfoot Grass. Chemical herbicides then become more unavoidable to deal with the affected yields. This research meets this challenge by using drone-based imaging along with Convolutional Neural Networks (CNNs) to detect and control weeds in real-time. It evaluates the performance of four CNN architectures in detecting these weeds within rice fields: ResNet101, DenseNet50, VGG19, and VGG16. Each model was trained to recognize unique weed types and minimize misclassifications for a high detection accuracy. Using metrics of precision, recall, and F1-score combined with confusion matrices to determine weed species classification capacity, the evaluation of the performances of the different models was executed. Amongst all the used models, it is found that ResNet101 gave the most accuracy with 98.2
In this paper, skin cancer is detected by a deep learning model using the EfficientNetB7 approach. The HAM10000 dataset was used to train and test the model. To sort out class imbalance issues in the dataset, data balancing techniques such as data augmentation are used for impartial analysis of training data. Using the EfficientNetB7 architecture, the proposed work investigates the application of deep learning methods for the automated identification and categorization of skin lesions. One of the most prevalent forms of cancer is skin cancer, and increasing treatment success rates and patient survival greatly depend on early detection. The main goal of the proposed work is to create a highly accurate and effective model that can precisely identify skin cancer, helping dermatologists with the diagnostic procedure. Training and evaluation have been conducted using the HAM10000 dataset, which is a varied collection of dermoscopic pictures of skin lesions. Data augmentation strategies have been used to address this problem and guarantee unbiased learning, which has improved the model’s generalization across various skin lesion categories. In order to improve performance, hyperparameter tuning and model optimization techniques have been used, with an accuracy of 92.04
