Sign in Subscribe
COVID-19

Enhancing Lung Disease Diagnosis with Deep-Learning-Based CT Scan Image Segmentation

Strategic Insights

— 1 min read

A new report from the Faculty of Engineering highlights the development of an enhanced computed tomography (CT) scan image segmentation method using deep learning, aimed at improving the diagnosis of lung diseases such as pneumonia, tuberculosis, bronchitis, emphysema, asthma, and COVID-19. The research proposes a preprocessing segmentation method using the UNet++ architecture, which has been tested on two datasets and showed significant improvements in accuracy, computational efficiency, and resource requirements. The study's findings suggest that the UNet++ L4 model is an ideal solution for accurate segmentation, computational efficiency, and affordable resource requirements.

Key Takeaways:

  • The research aims to enhance the efficiency of COVID-19 and other lung diseases diagnosis by developing a deep learning-based CT scan image segmentation method.
  • The proposed preprocessing segmentation method uses the UNet++ architecture, which has four levels (L1, L2, L3, and L4) and has been compared to several other models, including SegNet, FCANet, and DeepLabV3+.
  • Testing on two datasets, RSPHC and Kaggle, showed that the UNet++ L4 model achieved a Dice coefficient of 0.994 and 0.961, IoU of 0.989 and 0.930, computational time of 0.925 s and 1.189 s, and 9.16 million trainable parameters on both datasets.
  • The study concluded that the UNet++ L4 model is ideal for accurate segmentation, computational efficiency, and affordable resource requirements.
  • The research was conducted by Rima Tri Wahyuningrum and her team from the Department of Informatics Engineering, Faculty of Engineering, University of Trunojoyo Madura, Indonesia.

Statistics:

  • The UNet++ L4 model achieved a Dice coefficient of 0.994 and 0.961 on the RSPHC and Kaggle datasets, respectively.
  • The model achieved an IoU of 0.989 and 0.930 on the RSPHC and Kaggle datasets, respectively.
  • The computational time of the UNet++ L4 model was 0.925 s and 1.189 s on the RSPHC and Kaggle datasets, respectively.
  • The model had 9.16 million trainable parameters on both datasets.

Sources:

  • (Faculty of Engineering) Enhancing lung disease diagnosis with deep-learning-based CT scan image segmentation. Intelligent Systems with Applications, 2025,27():200565.
  • (NewsRx) New COVID-19 Study Findings Have Been Reported from Faculty of Engineering (Enhancing lung disease diagnosis with deep-learning-based CT scan image segmentation). Medical Imaging Week. September 13, 2025; p 142.

Read more