AI-Driven Pneumonia Detection with Interactive Decision Support

Deep learning isn’t just tweaking how respiratory diagnostics happen—it’s flipping the whole script. Medical professionals aren’t stuck squinting at X-rays by themselves anymore. Now, algorithms do the heavy lifting, running calculations and spotting patterns that might slip past even the keenest human eye[4]. In our study, we really put that to the test. We took MobileNetV2—a model already known for its efficiency—and tuned it using transfer learning[1], then threw in plenty of real-time data tweaks to make it even sharper[17]. The results? Almost 96.5% training accuracy, and the validation accuracy sat close behind at 94.2%. That tight gap tells you everything you need to know: the model’s not just parroting back the training data. It actually knows how to handle fresh, real-world cases.   What’s even more important is how accessible we’ve made this. Instead of demanding banks of high-powered GPUs or deep pockets for server time, our system runs just fine on a typical office computer[9]. A regular desktop in a local clinic? Good enough. We wanted to make sure clinics—and not just those in big, well-funded hospitals—could put this tool to work, even with limited technical staff or tight budgets.   But here’s the thing: accuracy sitting on a lab’s shelf doesn’t help patients or doctors in a busy ward. Deep learning models are notorious for being black boxes—sometimes they just spit out results, and nobody can say why. That’s not good enough when decisions involve a person’s life. To open up that black box, we built an interactive, web-based dashboard with a full suite of explainable AI tools[8]. For each X-ray, Grad-CAM visualizations kick in, literally spotlighting the murky spots on lung images[7]: regions that look extra cloudy, fluid build-ups, strange shadows that might signal something serious. Instead of dumping a probability score and calling it a day, the dashboard overlays clear, color-coded heatmaps right on the scan, along with confidence numbers. It’s evidence you can see—not just trust. Suddenly, doctors don’t have to guess why the AI said what it did. They get concrete visuals they can interrogate and discuss with colleagues or even with patients’ families.   We designed this system to help—not sideline—the experts. It’s not here to replace doctors, but to offer real, stress-relieving support when their workload piles up. The dashboard slides right into their routines, without forcing them to learn a dozen new workflows or deal with admin headaches. Radiologists spend less time hunched over screens and more time focusing on tricky cases. Fast triaging becomes possible, and standardized, easy-to-read PDF reports automatically sum up the findings, showing both the untouched X-ray and the highlighted suspicious spots for comparison. Since it’s all built on lightweight tech (just PHP and MySQL), it doesn’t care if it’s in a city hospital or a rural clinic with spotty internet—it fits right in, reliably and smoothly.   And we’re just scratching the surface. The way the system’s built, it’s ready to level up. With a bit more training, it can learn to tell apart not just pneumonia, but what kind—bacterial, viral, fungal, you name it. We know there’s a massive need to spot things like COVID-19, tuberculosis, and other tricky lung conditions. Our model’s architecture allows for these planned updates. Plus, there’s a clear path to move everything to a cloud-based web app and API, which opens doors for remote clinics or developing areas. When specialist doctors are hours—or even days—away, instant, trustworthy diagnostics from the cloud can genuinely change outcomes for patients.   In the end, this isn’t just about math or computer vision. It’s about creating tools that work for people: understandable, efficient, and ready to scale wherever they’re needed. We believe this is a crucial step toward a future where artificial intelligence empowers, rather than intimidates, both medical staff and the communities they serve. Whether it’s in big-city ERs or one-room clinics, this kind of AI doesn’t just promise better diagnostics; it makes them possible for everyone

  [1] F. Alshanketi, A. Alharbi, M. Kuruvilla, V. Mahzoon, S. T. Siddiqui, N. Rana, and A. Tahir, "Pneumonia detection from chest X-ray images using deep learning and transfer learning for imbalanced datasets," Journal of Imaging Informatics in Medicine, vol. 38, no. 1, 2024. [2] J. Colin and N. Surantha, "Interpretable deep learning for pneumonia detection using chest X-ray images," Information, vol. 16, no. 1, p. 53, 2025. [3] N. Darapaneni, A. Ranjan, D. Bright, D. Trivedi, K. Kumar, V. Kumar, and A. R. Paduri, "Pneumonia detection in chest X-rays using neural networks," arXiv preprint arXiv:2204.03618, 2022. [4] D. S. Kermany, M. Goldbaum, W. Zhang, and K. Zhang, "Identifying medical diagnoses and treatable diseases by image-based deep learning," Cell, vol. 172, no. 5, pp. 1122–1131, 2018. [5] I. D. Mienye, "Deep convolutional neural networks in medical image analysis: A review," Information, vol. 16, no. 3, p. 195, 2025. [6] M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L. C. Chen, "MobileNetV2: Inverted residuals and linear bottlenecks," in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 4510–4520, 2018. [7] R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, and D. Batra, "Grad-CAM: Visual explanations from deep networks via gradient-based localization," in Proceedings of the IEEE International Conference on Computer Vision (ICCV), pp. 618–626, 2017. [8] M. T. Zamir, "Explainable AI-driven analysis of radiology reports using text and image data: Experimental study," JMIR Formative Research, vol. 9, e77482, 2025. [9] M. S. A. Reshan, K. S. Gill, V. Anand, S. Gupta, H. Alshahrani, A. Sulaiman, and A. Shaikh, "Detection of pneumonia from chest X-ray images utilizing MobileNet model," Healthcare, vol. 11, no. 11, p. 1561, 2023. [10] D. Zhang, F. Ren, Y. Li, L. Na, and Y. Ma, "Pneumonia detection from chest X-ray images based on convolutional neural network," Electronics, vol. 10, no. 13, p. 1512, 2021.