The research team from the Department of Electronics and Communication Engineering has published a paper titled “RadiomixNet: Integrating Radiomics and Feature Extraction for Advanced Pneumonia Diagnosis” in the journal IEEE Access with an impact factor of 3.4. Prof. Siva Sankar Yellampalli, Professor of Practice, and Mr Rahul Gowtham, PhD Scholar, have worked on RadiomixNet, a smart computer-assisted system designed to help doctors diagnose pneumonia more accurately using chest X-ray images.

Abstract

The research presents RadiomixNet, a pneumonia diagnosis framework integrating radiomics-based feature extraction with advanced classification techniques. Chest X-ray images are pre-processed using denoising, resizing, and enhancement methods to ensure uniformity and high image quality. Radiomics features are extracted using Gray Level Co-Occurrence Matrix (GLCM), Gray Level Size Zone Matrix (GLSZM), Gray Level Run Length Matrix (GLRLM), and Gray Level Dependence Matrix (GLDM). Power Spectral Density (PSD) analysis using Burg, Yule Walker, and Welch techniques enhances the understanding of frequency characteristics within the radiomics feature matrices. To classify pneumonia cases, machine learning classifiers such as Bernoulli Naïve Bayes, Random Subspace Boost, Quadratic Discriminant, and Gradient Boosting are employed. Among these, Gradient Boosting demonstrated superior performance, achieving a Cohen’s Kappa of 0.93, MCC of 0.88, Youden’s Index of 0.82, and a Log Loss of 0.27. The proposed methodology enhances diagnostic accuracy, reduces variability in pneumonia detection, and provides a structured approach to feature-based pneumonia classification.

Explanation of the Research in Layperson’s Terms

Traditional diagnosis relies on a doctor visually examining the X-ray, which can sometimes lead to misinterpretations. RadiomixNet improves this process by using advanced image processing and artificial intelligence (AI) techniques.

1. Preprocessing the X-rays – Before analysis, we clean the images by removing noise (unwanted distortions), adjusting brightness, and resizing them to a standard format. This ensures all images are high quality and uniform.
2. Generating More Training Data – Since AI models need a large amount of data to learn effectively, we use Generative Adversarial Networks (GANs) to create additional synthetic X-ray images. This helps balance the dataset and improve the model’s ability to detect pneumonia accurately.
3. Extracting Hidden Patterns – The system breaks down X-ray images into tiny texture and shape details using advanced techniques like GLCM, GLSZM, GLRLM, and GLDM. These methods capture the structure of the lungs and highlight patterns that indicate pneumonia.
4. Analysing Frequency Components – Similar to how an audio equalizer separates different sound frequencies, we analyze the X-ray’s frequency components using techniques like Burg PSD, Yule Walker PSD, and Welch PSD. This helps uncover hidden details in the images that may not be visible to the human eye.
5. Making the Final Diagnosis – After extracting these detailed features, we use AI models to classify the images as “pneumonia” or “healthy.” We tested different models, including Naïve Bayes, Random Subspace Boost, Quadratic Discriminant, and Gradient Boosting. Among them, Gradient Boosting performed the best, making the most accurate predictions.
6. Evaluating Accuracy – To ensure the system is reliable, we used various accuracy-checking methods such as Cohen’s Kappa, Matthews Correlation Coefficient (MCC), Sensitivity, Specificity, Log Loss, and Brier Score.

Practical Implementation/Social Implications of the Research

Practical Implementation:

RadiomixNet has the potential to be integrated into real-world healthcare systems to assist in pneumonia diagnosis. Its implementation can take place in various ways:

1. Hospital Integration – RadiomixNet can be deployed in hospitals as a decision-support tool for radiologists. By analysing chest X-rays in real time, it can provide secondary validation, reducing diagnostic errors and improving accuracy in pneumonia detection.
2. Telemedicine and Remote Diagnosis – The system can be integrated into telemedicine platforms, allowing doctors in rural or under-resourced areas to diagnose pneumonia remotely. Patients can upload their X-ray images, and RadiomixNet can assist in providing a preliminary diagnosis.
3. Medical Imaging Centers – Radiology centers can incorporate RadiomixNet into their existing Picture Archiving and Communication Systems (PACS) to enhance diagnostic efficiency, reduce the workload of radiologists, and provide automated analysis.
4. Edge Computing in Low-Resource Settings – Unlike deep learning models that require expensive GPUs, RadiomixNet is optimized for standard computing hardware. This makes it feasible for implementation in clinics and hospitals that lack high-end computational resources.
5. Clinical Trials and Further Validation – Pilot studies in hospitals can validate RadiomixNet’s accuracy and reliability before widespread deployment. The system can be fine-tuned based on real-world patient data to improve its performance across diverse populations.

Social Implications:

1. Early and Accurate Diagnosis – By improving pneumonia detection, RadiomixNet can enable earlier treatment, reducing complications and mortality rates, especially in high-risk populations such as children, the elderly, and immunocompromised individuals.
2. Reducing Radiologist Workload – With increasing patient loads, radiologists often face diagnostic fatigue. RadiomixNet can act as an assistant, helping them focus on complex cases while automating routine pneumonia detection.
3. Bridging the Healthcare Gap – In developing countries where expert radiologists are scarce, RadiomixNet can assist general practitioners and healthcare workers in diagnosing pneumonia without requiring extensive radiology expertise.
4. Affordable and Scalable Solution – Since the system does not require expensive hardware, it can be implemented in low-resource settings, making advanced pneumonia detection accessible to a broader population.
5. Pandemic Preparedness – Pneumonia is a major complication of respiratory infections like COVID-19. RadiomixNet can be adapted to detect pneumonia-related lung infections, aiding in large-scale screening during outbreaks.

By integrating RadiomixNet into healthcare systems, we can enhance diagnostic accuracy, improve patient outcomes, and make pneumonia diagnosis more accessible and efficient globally.

Read more

RadiomixNet Implementation Framework

The Department of Electronics and Communication Engineering is proud to announce that Dr Durga Prakash M and his scholar Prasanthi Ms Prasanthi Lingala have their invention titled “An Organic Thin-Film Transistors (OTFTs) with Steep Subthreshold and Ultra-Low Temperature Solution Processing for Label-Free Biosensing” published in the Indian Patent Office Journal with the Application Number: 202541000088. Their research focus on developing an Organic Thin-Film Transistor (OTFT) that is able to work as a biosensor in detecting diseases or for real-time health monitoring.

Abstract

Organic Thin-Film Transistor (OTFT): The name “organic thin-film transistor” (OTFT) refers to a type of transistor that employs organic semiconductor materials in its active layer rather than the more traditional inorganic materials such as silicon. Optical thin-film transistors (OTFTs) are distinguished by their adaptability, low fabrication cost, and optimal applicability for electronic devices that are lightweight and portable. Considering their high sensitivity to changes in the surrounding environment and their compatibility with functionalised layers for the detection of biomolecules, these transistors find widespread application in the field of biosensors.

Explanation of the Research in Layperson’s Terms

Imagine a flexible electronic switch that can be bent, stretched, and used in lightweight devices—this is what an Organic Thin-Film Transistor (OTFT) does! Unlike traditional transistors made from rigid silicon, OTFTs use special organic materials, making them more adaptable for wearable sensors, flexible displays, and medical devices.

The research focuses on how these transistors can be used as biosensors, meaning they can detect tiny changes in the environment, like the presence of certain chemicals or biomolecules. This is important for medical testing, where OTFTs could help develop low-cost, highly sensitive diagnostic tools—imagine a simple patch that can detect diseases from sweat or a flexible sensor for real-time health monitoring! By improving how OTFTs interact with biological substances, the team aims to make them more accurate, efficient, and reliable for next-generation healthcare and wearable technology.

Fig.: Schematic structure of DNTT based OTFT