Comparative Analysis of Deep Learning Methods on CT Images for Lung Cancer Specification

KALKAN M., Guzel M. S., Ekinci F., SEZER E., Asuroglu T.

CANCERS, sa.19, 2024 (SCI-Expanded) identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Basım Tarihi: 2024
  • Doi Numarası: 10.3390/cancers16193321
  • Dergi Adı: CANCERS
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, BIOSIS, CAB Abstracts, CINAHL, EMBASE, Veterinary Science Database, Directory of Open Access Journals
  • Hacettepe Üniversitesi Adresli: Evet

Özet

Simple Summary Lung cancer is the most common form of cancer globally, making early detection essential for improving patient outcomes. In this study, deep learning methods were used to analyze CT lung images to detect cancer at an early stage. Several pre-trained models were employed, with InceptionResNetV2 achieving the highest detection accuracy of 98.5%. When cancer was detected, the extent of the tumor area was determined through segmentation models. The most accurate results were obtained using the InceptionUNet model, which achieved a Jaccard index of 95.3% in identifying the tumor area. These findings demonstrate the potential of deep learning for enhancing both the detection and localization of lung cancer, which may contribute to more effective treatment strategies.Abstract Background: Lung cancer is the leading cause of cancer-related deaths worldwide, ranking first in men and second in women. Due to its aggressive nature, early detection and accurate localization of tumors are crucial for improving patient outcomes. This study aims to apply advanced deep learning techniques to identify lung cancer in its early stages using CT scan images. Methods: Pre-trained convolutional neural networks (CNNs), including MobileNetV2, ResNet152V2, InceptionResNetV2, Xception, VGG-19, and InceptionV3, were used for lung cancer detection. Once the disease was identified, the tumor's region was segmented using models such as UNet, SegNet, and InceptionUNet. Results: The InceptionResNetV2 model achieved the highest detection accuracy of 98.5%, while UNet produced the best segmentation results, with a Jaccard index of 95.3%. Conclusions: The study demonstrates the effectiveness of deep learning models, particularly InceptionResNetV2 and UNet, in both detecting and segmenting lung cancer, showing significant potential for aiding early diagnosis and treatment. Future work could focus on refining these models and exploring their application in other medical domains.