Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy · 2026
Artificial intelligence-assisted FTIR spectroscopy for hormone receptor subtyping in formalin-fixed breast Cancer tissues.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
George Renee, Serrano Cherry Anne, Garong Jasmine Thea, Magsanay Kathryn, Maguigad Aleezah Claire, De Castro Maxene Alexandra, Navarro Andrea Noelle, Mandac Gabrielle Ricci, Ramos Kenaiah, Isais Maura, Abanador Peter Guillen, Gil Norely, Lenon Maria Sarah, Trinidad Marie Celestine, Baldomar Angeline Acryl, Tomas Rock Christian, Albano Pia Marie
Аннотация
BACKGROUND: Determination of estrogen receptor (ER) and progesterone receptor (PR) status is critical for breast cancer subtyping and guiding endocrine therapy. Although immunohistochemistry (IHC) remains the diagnostic gold standard, it is costly, labor-intensive, and prone to interobserver variability. These limitations are particularly restrictive in low-resource settings where access to standardized receptor testing is limited.
OBJECTIVE: This study presents a proof-of-concept evaluation of attenuated total reflectance-Fourier transform infrared (ATR-FTIR) spectroscopy combined with artificial intelligence (AI) for label-free classification of ER and PR status in formalin-fixed paraffin-embedded (FFPE) breast cancer tissues.
METHODS: A total of 72 samples (33 ER-positive, 39 ER-negative) were analyzed for ER classification, and 74 samples for PR classification (20 PR-positive, 54 PR-negative), generating 2328 and 1804 spectra, respectively. Spectra were acquired from pathologist-annotated tumor regions exhibiting definitive nuclear staining (positive) or absence thereof (negative) using a grid-based mapping strategy. Preprocessing included baseline correction (rubber-band algorithm) and z-score normalization. Seven AI models - logistic regression, support vector machine (SVM), decision tree, XGBoost, feedforward neural network (FNN), recurrent neural network (RNN), and convolutional neural network (CNN) - were trained and optimized using a genetic algorithm. Model performance was assessed via repeated cross-validation using AUC-ROC, accuracy, sensitivity, specificity, PPV, NPV, and F1 score.
RESULTS: CNN achieved the highest classification performance for both ER (AUC = 95.93% ± 6.64%, accuracy = 90.06% ± 4.85%) and PR (AUC = 97.46% ± 0.64%, accuracy = 91.51% ± 3.28%). FNN, RNN, and XGBoost also demonstrated strong performance, whereas SVM yielded the lowest accuracy and F1 scores. Statistically significant spectral differences between receptor-positive and -negative tumor regions were observed across biochemical bands corresponding to proteins, lipids, nucleic acids, and phosphorylated biomolecules.
CONCLUSION: AI-enhanced ATR-FTIR spectroscopy demonstrates high diagnostic potential for hormone receptor subtyping in FFPE tissues. As a label-free, scalable platform, it offers a promising alternative to IHC, particularly in resource-constrained environments. These findings establish the technical feasibility of this approach and warrant further validation in multicenter clinical cohorts.
OBJECTIVE: This study presents a proof-of-concept evaluation of attenuated total reflectance-Fourier transform infrared (ATR-FTIR) spectroscopy combined with artificial intelligence (AI) for label-free classification of ER and PR status in formalin-fixed paraffin-embedded (FFPE) breast cancer tissues.
METHODS: A total of 72 samples (33 ER-positive, 39 ER-negative) were analyzed for ER classification, and 74 samples for PR classification (20 PR-positive, 54 PR-negative), generating 2328 and 1804 spectra, respectively. Spectra were acquired from pathologist-annotated tumor regions exhibiting definitive nuclear staining (positive) or absence thereof (negative) using a grid-based mapping strategy. Preprocessing included baseline correction (rubber-band algorithm) and z-score normalization. Seven AI models - logistic regression, support vector machine (SVM), decision tree, XGBoost, feedforward neural network (FNN), recurrent neural network (RNN), and convolutional neural network (CNN) - were trained and optimized using a genetic algorithm. Model performance was assessed via repeated cross-validation using AUC-ROC, accuracy, sensitivity, specificity, PPV, NPV, and F1 score.
RESULTS: CNN achieved the highest classification performance for both ER (AUC = 95.93% ± 6.64%, accuracy = 90.06% ± 4.85%) and PR (AUC = 97.46% ± 0.64%, accuracy = 91.51% ± 3.28%). FNN, RNN, and XGBoost also demonstrated strong performance, whereas SVM yielded the lowest accuracy and F1 scores. Statistically significant spectral differences between receptor-positive and -negative tumor regions were observed across biochemical bands corresponding to proteins, lipids, nucleic acids, and phosphorylated biomolecules.
CONCLUSION: AI-enhanced ATR-FTIR spectroscopy demonstrates high diagnostic potential for hormone receptor subtyping in FFPE tissues. As a label-free, scalable platform, it offers a promising alternative to IHC, particularly in resource-constrained environments. These findings establish the technical feasibility of this approach and warrant further validation in multicenter clinical cohorts.