Lung Cancer Detection Using the Long Short-Term Memory(LSTM) Algorithm

سال انتشار: 1403
نوع سند: مقاله کنفرانسی
زبان: انگلیسی
مشاهده: 130

فایل این مقاله در 19 صفحه با فرمت PDF قابل دریافت می باشد

استخراج به نرم افزارهای پژوهشی:

لینک ثابت به این مقاله:

شناسه ملی سند علمی:

ICNABS01_052

تاریخ نمایه سازی: 15 بهمن 1403

چکیده مقاله:

Lung cancer is one of the most critical diseases in today's world. Despite being one of the most treatable forms of cancer, early detection is pivotal in reducing long-term mortality rates, as late-stage predictions significantly diminish survival probabilities. Automated disease diagnosis systems have emerged as a vital tool in the medical field, offering rapid responses, reliability, efficiency, and reduced mortality risks. Machine learning methods, in particular, have gained widespread popularity in detecting lung cancer, especially during its early stages. This study employs the Long Short-Term Memory (LSTM) algorithm, a supervised learning method, due to its ability to produce more stable and accurate models. Simulation results and comparisons with other studies demonstrate that utilizing the LSTM algorithm significantly enhances accuracy and sensitivity in lung cancer diagnosis. Specifically, our approach enables the identification of whether a suspicious region is a nodule. The evaluation indicates that the proposed method exhibits excellent sensitivity, especially in detecting small nodules. The estimated false positive rate of the proposed method is ۱۶.۶۷%, which is notably lower compared to other approaches. A qualitative comparison highlights the superiority of the proposed method in identifying small nodules, which even expert radiologists may find challenging to detect in imaging. Unlike other studies that primarily focus on larger nodules, our method prioritizes detecting smaller ones. Small nodule detection is inherently more complex due to their indistinct separation from other lung structures, which likely explains the superior performance of our approach. In experiments, the proposed method demonstrated robust performance in identifying nodules across the evaluated datasets. The efficiency of the proposed approach was assessed based on accuracy, sensitivity, and the Dice similarity coefficient. Results from the LIDC database indicate that the proposed method offers comparable accuracy and response times to existing approaches while providing distinct advantages in detecting complex nodules.

نویسندگان

Mahdi Kamalzare

Master of Artificial Intelligence, Islamic Azad University, Tehran Science and Research Branch, Iran

Saman Ghahraman

Master of Artificial Intelligence, Islamic Azad University, Tehran Science and Research Branch, Iran