EUVIMEDEUVIMED entdecken
Crossref

Optimizing Lung Segmentation In Chest X-Rays Through Comparative Preprocessing And Deep Learning Benchmarks

Ruchika Nagar, Dr. Rajendra N Solanki, Dr. Dipak Patel, Dr. Vibha Patel, Dr. Manish Patel

International Journal of Drug Delivery Technology · 11.05.2026 · 16 (26s)

DOI: 10.25258/ijddt.16.26s.31 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Drug Delivery Technology, online erschienen am 11.05.2026. Verfasst von Ruchika Nagar und 4 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This paper introduce a complete optimization process for lung segmentation in chest xrays (cxr) – a necessary step for a number of computer-assisted diagnoses, most notably tb and thoracic diseases. Here we compared two types of deep learning models to determine which type is better suited for cxr lung segmentation. We evaluated both a standard dense encoder-decoder network (with no attention mechanism) and a dense encoder-decoder network with an attention mechanism. Additionally, we studied nine different image enhancement approaches and found that a specific combination of preprocessing steps improved the performance of the deep neural networks. The nine preprocessing approaches included a variety of traditional histogram equalization techniques as well as hybrid approaches using the balance contrast enhancement technique (bcet) and discrete wavelet transform (dwt) to analyze the effectiveness of each approach, the we developed a total of six model configurations including a densenet201 unet model and an adaptive segunet model from existing papers. A thorough quantitative analysis of each model's performance utilized the intersection over union (iou), dice coefficient, and hausdorff distance metrics. Overall, we demonstrated that the use of a hybrid preprocessing approach (bcet+dwt) provides a significant increase in performance of deep encoder models by reaching a training dice coefficient of 0.9344 while also demonstrating that overly aggressive wavelet denoising negatively impacts attention-based models by eliminating vital textural information; this research demonstrates that a preprocessing stage optimized for cxr imaging combined with a back-end architecture utilizing densely interconnected layers is superior to attentiononly models for general lung segmentation applications and provides a cost-effective alternative.

Zitieren

APA

Nagar, R., Solanki, D. R. N., Patel, D. D., Patel, D. V., & Patel, D. M. (2026). Optimizing Lung Segmentation In Chest X-Rays Through Comparative Preprocessing And Deep Learning Benchmarks. International Journal of Drug Delivery Technology, 16(26s). https://doi.org/10.25258/ijddt.16.26s.31

Vancouver

Nagar R, Solanki DRN, Patel DD, Patel DV, Patel DM. Optimizing Lung Segmentation In Chest X-Rays Through Comparative Preprocessing And Deep Learning Benchmarks. International Journal of Drug Delivery Technology. 2026;16(26s). doi:10.25258/ijddt.16.26s.31

BibTeX
@article{Nagar2026,
  author = {Nagar, Ruchika and Solanki, Dr. Rajendra N and Patel, Dr. Dipak and Patel, Dr. Vibha and Patel, Dr. Manish},
  title = {Optimizing Lung Segmentation In Chest X-Rays Through Comparative Preprocessing And Deep Learning Benchmarks},
  journal = {International Journal of Drug Delivery Technology},
  year = {2026},
  volume = {16},
  number = {26s},
  doi = {10.25258/ijddt.16.26s.31}
}
Kostenlos starten →