Parvathaneni Naga Srinivasu

Dr. Parvathaneni Naga Srinivasu


Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 January 2025
Verified
Journal Article
Closed access
PDF file
PPPradeep Pradeep
Pradeep Pradeep
SPS. Phani Praveen
S. Phani Praveen
VTVipin Tiwari
Vipin Tiwari
PAPradeep Kumar Arya
Pradeep Kumar Arya
PSParvathaneni Naga Srinivasu
Parvathaneni Naga Srinivasu
ASAditi Sharma
Aditi Sharma
An inventive deep learning-based method for identifying financial fraud, revolutionizing e-commerce security in the process. The research offers a state-of-the-art setup that makes use of deep learning computations in the dynamic world of online exchanges, where the possibility of fraudulent activity is a danger. Since frauds are known to be erratic and lack consistency, it might be challenging to spot them. Con artists exploit the latest developments in technology. They manage to evade security measures, which results in millions of dollars being lost. One method of tracking fraudulent exchanges is to use information-mining techniques to investigate and detect unusual behaviours. Interactions. In contrast to deep learning techniques as auto encoders, convolutional neural networks (CNN), restricted Boltzmann machines (RBM), and deep belief networks (DBN), this paper aims to benchmark several machine-learning techniques, such as k-nearest neighbour (KNN), irregular forest, and support vector machines (SVM). The three-evaluation metrics that are really employed are the Area Under the ROC Curve (AUC), the Matthews Correlation Coefficient (MCC), and the Cost of Failure.

10 June 2026 in International Journal of Computational Intelligence Systems
Verified
Journal Article
Open access
PDF file
PSParvathaneni Naga Srinivasu
Parvathaneni Naga Srinivasu
ASAkshar Samudrala
Akshar Samudrala
PDPinapati Gagan Devesh
Pinapati Gagan Devesh
SRSurakanti Advaith Reddy
Surakanti Advaith Reddy
ANAhmed Al Nuaim
Ahmed Al Nuaim
Simplifying medical language is vital for effective healthcare communication, as patients often find it hard to understand complex clinical terms and lengthy, information-rich sentences. Recent advances in natural language processing (NLP) have led to powerful large language models(LLMs). However, their application in the medical domain remains constrained by substantial computational and hardware requirements. In this work, we investigate parameter-efficient fine-tuning strategies to adapt LLMs for simplifying clinical text while maintaining semantic fidelity and factual accuracy. Using Quantized Low-Rank Adaptation (QLoRA), we fine-tuned the Llama-3.1-8B-Instruct model on the Med-EASi and Cochrane datasets, achieving substantial gains in readability while preserving medical meaning. We evaluate multiple adapter designs, including Lightweight LoRA, TinyAttention, and standard LoRA with rank-4 decomposition, to analyze their effectiveness for domain-specific simplification tasks. Through 4-bit quantization of the base model combined with trainable low-rank adapters, the overall memory footprint is reduced by nearly an order of magnitude, enabling efficient fine-tuning on a single consumer-grade GPU. Our fine-tuned model achieves a Flesch-Kincaid grade level of 7.6 for MedEASI and 11.06 for the Cochrane dataset, along with an Automated Readability Index of 8.7 for MedEASI and 9.63 for the Cochrane dataset. Additionally, the model achieves BERTScore values of 0.845 and 0.864 on the MedEASI and Cochrane datasets, respectively, while improving reading-ease by +76.2 and +57.3 points over the source texts. Experimental results demonstrate improved simplification performance relative to selected baseline systems while preserving semantic similarity. These findings show that parameter-efficient fine-tuning methods can achieve performance comparable to full fine-tuning while significantly reducing computational cost. This efficiency supports deployment in resource-constrained healthcare settings and may improve patient access to complex clinical information.

23 October 2026 in Discover Artificial Intelligence
Verified
Journal Article
Open access
PDF file
HAHaritha Akkineni
Haritha Akkineni
MBMyneni Madhu Bala
Myneni Madhu Bala
DRDontha Madhusudhana Rao
Dontha Madhusudhana Rao
PSParvathaneni Naga Srinivasu
Parvathaneni Naga Srinivasu
Brain tumor subregion segmentation from multimodal magnetic resonance imaging (MRI) is important for glioma diagnosis, treatment planning, and disease monitoring. This study presents a ResNet50-U-Net integrated deep residual framework for segmenting Whole Tumor (WT), Tumor Core (TC), and Enhancing Tumor (ET) regions from multimodal MRI. Four MRI modalities, namely FLAIR, T1, T1ce, and T2, are combined as a four-channel input to transform the multimodal input compatible with the ResNet50 encoder. The proposed ResNet50-U-Net was compared with CNN and U-Net baselines using Dice Similarity Coefficient, Intersection over Union, HD95, precision, sensitivity, specificity, accuracy, parameter count, model size, and training time. The primary experiments were conducted on 368 labelled BraTS2020 cases using a patient-wise split of 257 training, 55 validation, and 56 independent test cases. Over tumor slice selection, 17,053 for training, 3,573 for validation, and 3,706 for testing were obtained. The proposed ResNet50-U-Net achieved the highest Dice scores for WT and ET segmentation, with values of 0.9100 and 0.8151, respectively, while the CNN achieved the highest TC Dice score of 0.7915. Cross-dataset generalization was evaluated on 95 independent BraTS-Africa glioma cases using an independently retrained instance of the same ResNet50-U-Net architecture, without fine-tuning or external threshold optimization. It is observed that patient-level Dice scores of 0.8592 ± 0.1674, 0.7387 ± 0.2500, and 0.6990 ± 0.2310 were obtained for WT, TC, and ET, respectively. These findings indicate that the proposed framework provides strong WT and ET segmentation on BraTS2020 and retains meaningful segmentation capability on an unseen external cohort, although cross-dataset robustness varies across tumor subregions.