928P - Radiomic analysis based on machine learning of multi-MR sequences to assess early treatment response in locally advanced nasopharyngeal carcinoma

Background

The prediction of early response in locally advanced nasopharyngeal carcinoma (LA-NPC) after concurrent chemoradiotherapy (CCRT) is indeed important for determining the need for timely consolidation therapy. Radiomics has helped medical imaging expand from diagnostic to a personalized clinical decision aid, while most traditional radiomics primarily focus on mono-sequence. Recently, multi-MR sequences model has shown better outcomes in the management of many cancers. However, the validation of a radiomic analysis of multi-MR sequences based on machine learning for this purpose is currently lacking for patients with LA-NPC after CCRT.

Methods

This study comprised 104 patients with LA-NPC, of which 70% were allocated for training and the remaining 30% were used for internal validation. We extracted radiomics features corresponding to each mono-sequences (T1, T1C, T2, DWI, and ADC), and then applied LASSO for feature selection and constructed the risk model using machine learning algorithms including SVM, Random Forest, ExtraTrees, XGBoost and LightGBM. Performance comparisons were made between mono-sequences and then feature fusion was performed to construct a multi-MR sequences model. Clinical features were numerically mapped and analyzed using machine learning algorithms similar to those in the Radiomics Model. Finally, we created a combined model that integrated the radiomics and clinical models to enhance clinical relevance, whose diagnostic effectiveness was evaluated using receiver operator characteristic (ROC) curves in the test cohort.

Results

In the training cohort, while the Clinic model achieved an Area Under the Curve (AUC) of 0.892 and the Fusion model reached 0.972, the Combined model excelled with an AUC of 0.990. Similarly, in the test cohort, the Combined model outperformed both the Clinic (AUC: 0.852) and Fusion models (AUC: 0.886), achieving an AUC of 0.900.

Conclusions

As a result, a combined model, which merges clinical and multi-MR sequences radiomics model, showed good performance for predicting early response of LA-NPC after CCRT.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

Jiangsu Province Entrepreneurship and Innovation Doctoral Talent Program and Jiangsu Province People's Hospital Clinical Capability Enhancement Project.

Disclosure

All authors have declared no conflicts of interest.