WCLC 2025 - Posters & ePosters-P2.08 .30 Ia-Based Prediction in NSCLC Patients Undergoing Neoadjuvant Radiochemotherapy

Pdf Summary

This study addresses the challenge of predicting major pathological response (MPR) in patients with non-small cell lung cancer (NSCLC) undergoing neoadjuvant therapy. Surgical cure is limited to 20–30% of NSCLC patients, and high recurrence rates persist due to micrometastases. Achieving pathologic response, especially major pathologic response (MPR) or pathologic complete response (pCR), is linked to better long-term survival, making early prediction valuable for tailoring treatment plans.<br /><br />The authors developed a multimodal AI framework called Doctor-in-the-Loop that integrates pre-treatment CT imaging and clinical data, both pre- and peri-treatment, to non-invasively predict MPR. The model employs a progressive training strategy to focus learning on clinically relevant regions via expert-defined lung and tumor masks, enhanced with an explainability loss that aligns model attention with these areas using heatmaps. Five models were trained encompassing imaging only, clinical data before treatment, clinical data during treatment, and combined multimodal models pre- and on-treatment.<br /><br />The multimodal model integrating on-treatment clinical information with imaging achieved the highest predictive performance, emphasizing the importance of incorporating treatment-phase data. This approach holds promise for guiding personalized neoadjuvant therapy in NSCLC by early identification of responders, potentially reducing unnecessary toxicity, enabling treatment escalation, or sparing surgery in select patients.<br /><br />The study analyzed 100 NSCLC patients treated at Campus Bio-Medico University Hospital Rome from 2005–2022, considering clinical factors such as age, gender, stage, smoking history, ECOG status, radiotherapy technique, surgery type, and histology.<br /><br />In summary, the Doctor-in-the-Loop explainable AI model leverages multimodal data with focused attention mechanisms to predict pathological response in NSCLC patients before and during treatment, supporting improved individualized treatment strategies in the neoadjuvant setting.

Asset Subtitle

Claudia Tacconi

Meta Tag

Speaker Claudia Tacconi

Topic Local-Regional Non-small Cell Lung Cancer

Keywords

non-small cell lung cancer

major pathological response

neoadjuvant therapy

Doctor-in-the-Loop AI framework

multimodal data integration

CT imaging

clinical data

progressive training strategy

explainability loss

personalized treatment prediction