Department of Electrical and Electronics Engineering, John Cox Memorial CSI Institute of Technology, Thiruvananthapuram, India.
Department of Computer Science and Engineering, Baselios Mathews II College of Engineering, Kollam, Kerala, India.
Curr Med Imaging. 2022;18(12):1282-1290. doi: 10.2174/1573405617666211126151713.
As the mortality rate of lung cancer is enormously high, its impact is also extremely higher than the other types of cancer. Lung malignancy is thus considered one of the deadliest diseases with a high death rate in the world. It is reported that nearly 1.2 million people are diagnosed with this disease and about 1.1 million individuals are died due to this type of cancer every year. The early detection of this disease is the only solution for minimizing the death rate or maximizing the survival rate. However, the timely identification of lung malignant growth is a complex process and hence various imaging algorithms are employed in the process of detecting lung cancer on time.
The Computer-Aided Diagnosis (CAD) is highly beneficial for the radiologist to rapidly detect and diagnose the irregularities in advance. The CAD systems usually focus on identifying and detecting the lung nodules. As the treatment of this disease is provided on the basis of its stages, the early detection of cancer has to be given much importance. The major drawbacks of existing CAD systems are less accuracy in segmenting the nodule and staging the lung cancer.
The major aim of this work is to categorize the lung nodules from the CT image and classify the tumorous cells for identifying the exact position of cancer with higher sensitivity, precision, and accuracy than other strategies.
The methods employed in this study are listed as follows: (i) For the process of de-noising and edge sharpening of lung image, the curvelet transform was used. (ii) The Fuzzy thresholding technique was used to perform lung image binarization and lung boundary corrections. (iii) Segmentation was performed by implementing the K-means algorithm. (iv) By using Convolutional Neural Network (CNN), different stages of lung nodules, like benign and malignant, were identified.
The proposed classifier achieves optimal accuracy of 97.3%, a sensitivity of 98.6% and a specificity of 96.1% which are significantly better than the other approaches. Thus, the proposed approach is highly helpful in detecting lung cancer in its early stages.
The results validate that the proposed algorithms are highly capable of classifying the lung images into various stages, which effectively helps the radiologist in the decision-making process.
由于肺癌的死亡率极高，其影响也远远超过其他类型的癌症。因此，肺癌被认为是世界上最致命的疾病之一，死亡率很高。据报道，每年有近 120 万人被诊断出患有这种疾病，约有 110 万人因此类癌症死亡。这种疾病的早期发现是降低死亡率或提高生存率的唯一途径。然而，及时发现肺部恶性生长是一个复杂的过程，因此在及时发现肺癌的过程中采用了各种成像算法。
计算机辅助诊断（CAD）对放射科医生快速发现和提前诊断非常有益。CAD 系统通常专注于识别和检测肺结节。由于这种疾病的治疗是基于其分期的，因此必须高度重视癌症的早期发现。现有 CAD 系统的主要缺点是结节分割和肺癌分期的准确性较低。
这项工作的主要目的是从 CT 图像中对肺结节进行分类，并对肿瘤细胞进行分类，以比其他策略更高的灵敏度、精度和准确性来识别癌症的确切位置。
本研究采用的方法如下：（i）对肺图像进行去噪和边缘锐化，采用曲线波变换。（ii）采用模糊阈值技术进行肺图像二值化和肺边界修正。（iii）采用 K-均值算法进行分割。（iv）利用卷积神经网络（CNN）识别肺结节的不同阶段，如良性和恶性。
所提出的分类器达到了 97.3%的最佳精度、98.6%的灵敏度和 96.1%的特异性，明显优于其他方法。因此，该方法有助于早期发现肺癌。
结果验证了所提出的算法能够将肺图像分类到各个阶段，这有效地帮助放射科医生进行决策。
