Multimodal Data Integration via Multilayer Perceptron for Real-Time Intraoperative Discrimination of Hepatocellular Carcinoma and Non-Cancerous Liver Tissue

Erol Karaaslan; Emek Güldoğan; Mehmet Cengiz Çolak

Makale detayları

Multimodal Data Integration via Multilayer Perceptron for Real-Time Intraoperative Discrimination of Hepatocellular Carcinoma and Non-Cancerous Liver Tissue

Yazarlar:

Erol Karaaslan | Emek Güldoğan | Mehmet Cengiz Çolak

Yayıncı:

Ordu University

Ana Başlıklar:

Gastroenterology and Hepatology

Dergi:

Middle Black Sea Journal of Health Science Vol.11 Issue 3, pp.212-227

Belge Tipi:

Makale

Yayın Tarihi:

8/2025

Belge Dili:

İngilizce

Document ID (Org No.):

2001357760

pISSN:

2149-7796

Anahtar kelimeler:

Özet:

 Objective: Hepatocellular carcinoma HCC remains a leading cause of cancer-related mortality, necessitating advanced diagnostic tools to improve surgical outcomes. This study presents a novel multilayer perceptron MLP neural network model trained on the Multimodal Liver Cancer Surgical Dataset MLCSD, integrating spectroscopic, hemodynamic, and biochemical features to distinguish HCC from non-cancerous tissue. Method: The dataset comprises 3,243 patient records, including tissue endogenous fluorescence TEF, laser Doppler flowmetry LDF signals, and biomarkers AFP, albumin, bilirubin. Following standardization and a 70:30 train-test split, the MLP was constructed for the related task. Different metrics were used for model evaluation. Results: The MLP modelachieved exceptional accuracy 97.8 test accuracy, 99.0 training accuracy and near-perfect discrimination AUC  0.999. Key predictors included albumin normalized importance: 100, respiratory-driven LDF LDF_Resp, 91, and AFP 66, underscoring their biological relevance. Secondary contributors e.g., spectral kurtosis, tissue density added marginal value, suggesting potential for streamlined biomarker panels. The models high sensitivity 95.8 and specificity 98.8 highlight its potential for real-time intraoperative tissue classification, enabling precise tumour margin delineation and preservation of functional hepatic parenchymaa critical advantage for cirrhotic patients. Conclusion: Despite promising results, limitations include reliance on simulated clinical data and binary classification without longitudinal endpoints e.g., recurrence risk, microvascular invasion. Future work must validate the model in prospective, multi-institutional cohorts and integrate genomic and radiomics data for comprehensive prognostication. This study demonstrates that AI-driven multimodal integration significantly outperforms conventional unimodal approaches, offering a foundation for personalized surgical navigation systems in HCC management. The findings advocate for further research into whole-patient AI models combining real-time physiological signals with molecular and imaging biomarkers to optimize oncological outcomes.

 Objective: Hepatocellular carcinoma HCC remains a leading cause of cancer-related mortality, necessitating advanced diagnostic tools to improve surgical outcomes. This study presents a novel multilayer perceptron MLP neural network model trained on the Multimodal Liver Cancer Surgical Dataset MLCSD, integrating spectroscopic, hemodynamic, and biochemical features to distinguish HCC from non-cancerous tissue. Method: The dataset comprises 3,243 patient records, including tissue endogenous fluorescence TEF, laser Doppler flowmetry LDF signals, and biomarkers AFP, albumin, bilirubin. Following standardization and a 70:30 train-test split, the MLP was constructed for the related task. Different metrics were used for model evaluation. Results: The MLP modelachieved exceptional accuracy 97.8 test accuracy, 99.0 training accuracy and near-perfect discrimination AUC  0.999. Key predictors included albumin normalized importance: 100, respiratory-driven LDF LDF_Resp, 91, and AFP 66, underscoring their biological relevance. Secondary contributors e.g., spectral kurtosis, tissue density added marginal value, suggesting potential for streamlined biomarker panels. The models high sensitivity 95.8 and specificity 98.8 highlight its potential for real-time intraoperative tissue classification, enabling precise tumour margin delineation and preservation of functional hepatic parenchymaa critical advantage for cirrhotic patients. Conclusion: Despite promising results, limitations include reliance on simulated clinical data and binary classification without longitudinal endpoints e.g., recurrence risk, microvascular invasion. Future work must validate the model in prospective, multi-institutional cohorts and integrate genomic and radiomics data for comprehensive prognostication. This study demonstrates that AI-driven multimodal integration significantly outperforms conventional unimodal approaches, offering a foundation for personalized surgical navigation systems in HCC management. The findings advocate for further research into whole-patient AI models combining real-time physiological signals with molecular and imaging biomarkers to optimize oncological outcomes.

 Objective: Hepatocellular carcinoma HCC remains a leading cause of cancer-related mortality, necessitating advanced diagnostic tools to improve surgical outcomes. This study presents a novel multilayer perceptron MLP neural network model trained on the Multimodal Liver Cancer Surgical Dataset MLCSD, integrating spectroscopic, hemodynamic, and biochemical features to distinguish HCC from non-cancerous tissue. Method: The dataset comprises 3,243 patient records, including tissue endogenous fluorescence TEF, laser Doppler flowmetry LDF signals, and biomarkers AFP, albumin, bilirubin. Following standardization and a 70:30 train-test split, the MLP was constructed for the related task. Different metrics were used for model evaluation. Results: The MLP modelachieved exceptional accuracy 97.8 test accuracy, 99.0 training accuracy and near-perfect discrimination AUC  0.999. Key predictors included albumin normalized importance: 100, respiratory-driven LDF LDF_Resp, 91, and AFP 66, underscoring their biological relevance. Secondary contributors e.g., spectral kurtosis, tissue density added marginal value, suggesting potential for streamlined biomarker panels. The models high sensitivity 95.8 and specificity 98.8 highlight its potential for real-time intraoperative tissue classification, enabling precise tumour margin delineation and preservation of functional hepatic parenchymaa critical advantage for cirrhotic patients. Conclusion: Despite promising results, limitations include reliance on simulated clinical data and binary classification without longitudinal endpoints e.g., recurrence risk, microvascular invasion. Future work must validate the model in prospective, multi-institutional cohorts and integrate genomic and radiomics data for comprehensive prognostication. This study demonstrates that AI-driven multimodal integration significantly outperforms conventional unimodal approaches, offering a foundation for personalized surgical navigation systems in HCC management. The findings advocate for further research into whole-patient AI models combining real-time physiological signals with molecular and imaging biomarkers to optimize oncological outcomes.

 Objective: Hepatocellular carcinoma HCC remains a leading cause of cancer-related mortality, necessitating advanced diagnostic tools to improve surgical outcomes. This study presents a novel multilayer perceptron MLP neural network model trained on the Multimodal Liver Cancer Surgical Dataset MLCSD, integrating spectroscopic, hemodynamic, and biochemical features to distinguish HCC from non-cancerous tissue. Method: The dataset comprises 3,243 patient records, including tissue endogenous fluorescence TEF, laser Doppler flowmetry LDF signals, and biomarkers AFP, albumin, bilirubin. Following standardization and a 70:30 train-test split, the MLP was constructed for the related task. Different metrics were used for model evaluation. Results: The MLP modelachieved exceptional accuracy 97.8 test accuracy, 99.0 training accuracy and near-perfect discrimination AUC  0.999. Key predictors included albumin normalized importance: 100, respiratory-driven LDF LDF_Resp, 91, and AFP 66, underscoring their biological relevance. Secondary contributors e.g., spectral kurtosis, tissue density added marginal value, suggesting potential for streamlined biomarker panels. The models high sensitivity 95.8 and specificity 98.8 highlight its potential for real-time intraoperative tissue classification, enabling precise tumour margin delineation and preservation of functional hepatic parenchymaa critical advantage for cirrhotic patients. Conclusion: Despite promising results, limitations include reliance on simulated clinical data and binary classification without longitudinal endpoints e.g., recurrence risk, microvascular invasion. Future work must validate the model in prospective, multi-institutional cohorts and integrate genomic and radiomics data for comprehensive prognostication. This study demonstrates that AI-driven multimodal integration significantly outperforms conventional unimodal approaches, offering a foundation for personalized surgical navigation systems in HCC management. The findings advocate for further research into whole-patient AI models combining real-time physiological signals with molecular and imaging biomarkers to optimize oncological outcomes.

Makale detayları

Multimodal Data Integration via Multilayer Perceptron for Real-Time Intraoperative Discrimination of Hepatocellular Carcinoma and Non-Cancerous Liver Tissue

Yazarlar:

Erol Karaaslan | Emek Güldoğan | Mehmet Cengiz Çolak

Yayıncı:

Ordu University

Ana Başlıklar:

Gastroenterology and Hepatology

Dergi:

Middle Black Sea Journal of Health Science Vol.11 Issue 3, pp.212-227

Belge Tipi:

Makale

Yayın Tarihi:

8/2025

Belge Dili:

İngilizce

Document ID (Org No.):

2001357760

pISSN:

2149-7796

Hepatocellular Carcinoma

Artificial Intelligence

Multimodal Data Integration

Machine Learning

Multilayer Perceptron.

Sayı

Makale detayları

Multimodal Data Integration via Multilayer Perceptron for Real-Time Intraoperative Discrimination of Hepatocellular Carcinoma and Non-Cancerous Liver Tissue

Yazarlar: Erol Karaaslan | Emek Güldoğan | Mehmet Cengiz Çolak

Yayıncı: Ordu University

Ana Başlıklar: Gastroenterology and Hepatology

Dergi: Middle Black Sea Journal of Health Science Vol.11 Issue 3, pp.212-227

Belge Tipi: Makale

Yayın Tarihi: 8/2025

Belge Dili: İngilizce

Document ID (Org No.): 2001357760

pISSN: 2149-7796

Hepatocellular Carcinoma

Artificial Intelligence

Multimodal Data Integration

Machine Learning

Multilayer Perceptron.

Sayı

Yazarlar:

Erol Karaaslan | Emek Güldoğan | Mehmet Cengiz Çolak

Yayıncı:

Ordu University

Ana Başlıklar:

Gastroenterology and Hepatology

Dergi:

Middle Black Sea Journal of Health Science Vol.11 Issue 3, pp.212-227

Belge Tipi:

Makale

Yayın Tarihi:

8/2025

Belge Dili:

İngilizce

Document ID (Org No.):

2001357760

pISSN:

2149-7796