A Machine Learning Approach to Keratin Filament Imaging: Enhancing Risk Stratification for Skin Cancer Metastasis
Bisong Ko, Eunsu Lee, Chengmin Sun, Zai Lee, Jisu Park, Yuna Jung, Beonjun Kim, Seunghyun Cho, Ikon Kim, Minseo Ha
iGEM-TJI-Seoul, Seoul, South Korea
Publication date: May 31, 2025
iGEM-TJI-Seoul, Seoul, South Korea
Publication date: May 31, 2025
DOI: http://doi.org/10.34614/JIYRC2025I17
ABSTRACT
Skin cancer metastasis is a process that includes cytoskeleton rearrangements that enable cancer cells to migrate and invade other tissues. Keratin intermediate filaments are essential structural components of the cytoskeleton of keratinocytes and are involved in cell survival and metastatic potential. Prior research has shown that changes in the structure of keratin filaments are linked with higher malignancy and worse prognosis in epithelial tumors. This work describes a new method of risk assessment of metastatic skin cancer based on deep learning and visualization of keratin intermediate filaments. The model then compares the structural parameters of the filaments, including length and orientation, to determine the specific variations in cytoskeletal architecture between the fast and slow-moving keratinocytes and relates these features to metastatic ability. Immunofluorescent images were analyzed at high magnification and processed using ImageJ. A convolutional neural network (CNN) model created through Google Teachable Machine yielded 85% accuracy in differentiating between high and low metastatic risk cells. This non-invasive approach minimizes histopathological examination and invasive biopsy and provides an objective diagnostic technique. The results indicate that keratin filament features could be promising biomarkers of cancer progression in clinical oncology to facilitate early risk assessment and individually tailored treatment.
Skin cancer metastasis is a process that includes cytoskeleton rearrangements that enable cancer cells to migrate and invade other tissues. Keratin intermediate filaments are essential structural components of the cytoskeleton of keratinocytes and are involved in cell survival and metastatic potential. Prior research has shown that changes in the structure of keratin filaments are linked with higher malignancy and worse prognosis in epithelial tumors. This work describes a new method of risk assessment of metastatic skin cancer based on deep learning and visualization of keratin intermediate filaments. The model then compares the structural parameters of the filaments, including length and orientation, to determine the specific variations in cytoskeletal architecture between the fast and slow-moving keratinocytes and relates these features to metastatic ability. Immunofluorescent images were analyzed at high magnification and processed using ImageJ. A convolutional neural network (CNN) model created through Google Teachable Machine yielded 85% accuracy in differentiating between high and low metastatic risk cells. This non-invasive approach minimizes histopathological examination and invasive biopsy and provides an objective diagnostic technique. The results indicate that keratin filament features could be promising biomarkers of cancer progression in clinical oncology to facilitate early risk assessment and individually tailored treatment.
