Related Experiment Videos
MILGDF: a multi-task, instance-level supervised model for oral squamous cell carcinoma integrating local-global
Chen-Xi Li1,2,3, Yan Chen4, Liang-Hui Xu5
1Jiangsu Province Engineering Research Center of Stomatological Translational Medicine, Affiliated Stomatological Hospital of Nanjing Medical University, Nanjing, People's Republic of China lichenxi@xjmu.edu.cn lichenxiuke@gmail.com.
Aims:
This research was designed to establish an innovative diagnostic strategy employing whole-slide imaging (WSI) technology to address the diagnostic difficulties arising from the intricate histological architecture and morphological diversity observed in oral squamous cell carcinoma (OSCC). The developed methodology enables precise early identification and histomorphology-driven prognostic stratification of malignant lesions, thereby improving clinical management and patient prognosis.
Methods:
We propose a multi-task learning framework that combines local-global attention mechanisms with adaptive decision fusion (MILGDF). This model utilises instance-level category-specific attention to enhance feature extraction efficacy while overcoming the limitations inherent in traditional bag-level attention methods. An adaptive weighting system was incorporated to dynamically adjust the contribution of local and global features, ensuring optimal performance in dual tasks of OSCC diagnosis and prognostic stratification.
Results:
Rigorous validation on the HIDOC and TCGA-OSCC datasets revealed the predictive performance of our model. The MILGDF framework attained an area under the curve of 0.952 (accuracy: 0.909) on HIDOC and 0.745 (accuracy: 0.725) on TCGA-OSCC. Statistical comparison using DeLong's test and paired t-tests demonstrated significantly superior performance (p<0.05) over existing comparative models in both diagnostic classification and prognostic stratification.
Conclusions:
Our findings demonstrate that the MILGDF model represents an improvement in whole-slide image-based OSCC analysis, delivering enhanced diagnostic accuracy and prognostic reliability relative to current approaches. The framework's consistent performance highlights its computational benchmark value and preliminary translational potential for early OSCC detection and auxiliary treatment planning, serving as a valuable asset for prognostic evaluation and therapeutic strategy formulation.