Fallahzadeh Omid, Dehghani-Bidgoli Zohreh, Assarian Mohammad
Department of Electrical and Computer Engineering, Kashan Branch, Islamic Azad University, Kashan, Iran.
Lasers Med Sci. 2018 Nov;33(8):1799-1806. doi: 10.1007/s10103-018-2544-3. Epub 2018 Jun 4.
Pathology as a common diagnostic test of cancer is an invasive, time-consuming, and partially subjective method. Therefore, optical techniques, especially Raman spectroscopy, have attracted the attention of cancer diagnosis researchers. However, as Raman spectra contain numerous peaks involved in molecular bounds of the sample, finding the best features related to cancerous changes can improve the accuracy of diagnosis in this method. The present research attempted to improve the power of Raman-based cancer diagnosis by finding the best Raman features using the ACO algorithm. In the present research, 49 spectra were measured from normal, benign, and cancerous breast tissue samples using a 785-nm micro-Raman system. After preprocessing for removal of noise and background fluorescence, the intensity of 12 important Raman bands of the biological samples was extracted as features of each spectrum. Then, the ACO algorithm was applied to find the optimum features for diagnosis. As the results demonstrated, by selecting five features, the classification accuracy of the normal, benign, and cancerous groups increased by 14% and reached 87.7%. ACO feature selection can improve the diagnostic accuracy of Raman-based diagnostic models. In the present study, features corresponding to ν(C-C) αhelix proline, valine (910-940), νs(C-C) skeletal lipids (1110-1130), and δ(CH2)/δ(CH3) proteins (1445-1460) were selected as the best features in cancer diagnosis.
病理学作为癌症的一种常见诊断测试，是一种侵入性、耗时且部分主观的方法。因此，光学技术，尤其是拉曼光谱，已引起癌症诊断研究人员的关注。然而，由于拉曼光谱包含与样品分子键相关的众多峰，找到与癌变变化相关的最佳特征可以提高该方法的诊断准确性。本研究试图通过使用蚁群优化（ACO）算法找到最佳拉曼特征来提高基于拉曼的癌症诊断能力。在本研究中，使用785纳米的显微拉曼系统从正常、良性和癌性乳腺组织样本中测量了49个光谱。在进行去除噪声和背景荧光的预处理后，提取生物样本12个重要拉曼带的强度作为每个光谱的特征。然后，应用蚁群优化算法找到诊断的最佳特征。结果表明，通过选择五个特征，正常、良性和癌性组的分类准确率提高了14%，达到87.7%。蚁群优化特征选择可以提高基于拉曼的诊断模型的诊断准确性。在本研究中，对应于ν（C-C）α螺旋脯氨酸、缬氨酸（910 - 940）、νs（C-C）骨架脂质（1110 - 1130）和δ（CH2）/δ（CH3）蛋白质（1445 - 1460）的特征被选为癌症诊断中的最佳特征。
