A multimodal scanning electron microscopy-atomic force microscopy images for early cervical cancer detection using discrete wavelet transform-derived features and divergence-based classification.

objectiveThe aim of this study was to enhance the early diagnosis of cervical cancer by utilizing both scanning atomic force microscopy and scanning electron microscopy images. Until now, these two imaging modalities have not been combined for this purpose. Relying on a single imaging technique may reduce diagnostic reliability. Therefore, a multimodal approach was developed to provide more accurate and consistent results.

methodsA total of 540 scanning electron microscopy and 540 atomic force microscopy cell images were obtained from the Faculty of Nanotechnology and Scanning Electron Laboratories. The discrete wavelet transform method was applied to extract ideal feature regions from the images. At least three divergence-based classifiers, triangle divergence, Jensen-Shannon divergence, and Hellinger divergence, were employed to identify cervical cancer-related patterns. To improve classification reliability, feature likelihoods were calculated using exponential graph regulation. Feature weights were determined and incorporated into a combined classification function. Based on this, scanning electron microscopy and atomic force microscopy cells were categorized into six groups: normal scanning electron microscopy, normal atomic force microscopy, benign scanning electron microscopy, benign atomic force microscopy, and malignant scanning electron microscopy, malignant atomic force microscopy.

resultsThe classifiers demonstrated interrelated performance, with triangle divergence achieving superior accuracy compared to Jensen-Shannon divergence and Hellinger divergence. The weighted feature approach improved diagnostic precision. The proposed method successfully predicted the likelihood of cervical cancer by combining scanning electron microscopy and atomic force microscopy data.

conclusionThe findings indicate that the triangle divergence-based multimodal classification outperforms single-modality approaches. The combined scanning electron microscopy and atomic force microscopy method using triangle divergence achieved a classification accuracy of 95.9%, demonstrating the effectiveness of this integrated approach for early cervical cancer detection.