Identifications of Lung Cancer Using Kernel Weighted Fuzzy Local Information C-Means Algorithm

• Published in: Traitement du signal Vol. 42; no. 2; pp. 1173 - 1184
• Main Authors: Karthikeyan, Sampathkumar, Kalaiselvi, Subbarayan, Sheela, John Bosco Joselin Jeya, Ashok, Maram
• Format: Journal Article
• Language: English
• Published: Edmonton International Information and Engineering Technology Association (IIETA) 01.04.2025
• Subjects: Algorithms; Annotations; Classification; Damping; Data points; Datasets; Identification; Image databases; Image segmentation; Lung cancer; Machine learning; Medical diagnosis; Medical imaging; Medical screening; Parameters; Pixels; Tradeoffs
• ISSN: 0765-0019; 1958-5608; 1958-5608
• DOI: 10.18280/ts.420247

An improved version of the Fuzzy C-Means (FCM) method called Kernel Weighted Fuzzy Local Information C-Means (KWFLICM), which incorporates a Kernel Distance Measure (KDM), and a trade-off Weighted Fuzzy Factor (WFF) for image segmentation is proposed. The WFF considers spatial distance and the intensity difference of all pixels in the surrounding area simultaneously. The KWFLICM algorithm uses WFF to precisely determine the damping extent of pixels next to one another. The target function is improved by adding KDM, making it even more robust to noise and outliers. Adaptive kernel parameters are determined using an efficient bandwidth selection mechanism. The distance variance of each data point is used to calculate these parameters via a process of comparison. The KDM and the parameter-free WFF trade-off improve the segmentation accuracy of the KWFLICM algorithm. Simulation results on actual and simulated images show that the KWFLICM algorithm performs well against noisy images. KWFLICM’s combination of kernel mapping and spatial weighting enables it to produce better segmentation and classification results in lung cancer identification. The KWFLICM algorithm’s noise resilience, accurate boundary detection, and sensitivity to small or complex tumor structures make it especially valuable in lung cancer detection on two benchmark databases, including LIDC and ELCAP.