Download This Paper
Preprints with The Lancet is a collaboration between The Lancet Group of journals and SSRN to facilitate the open sharing of preprints for early engagement, community comment, and collaboration. Preprints available here are not Lancet publications or necessarily under review with a Lancet journal. These preprints are early-stage research papers that have not been peer-reviewed. The usual SSRN checks and a Lancet-specific check for appropriateness and transparency have been applied. The findings should not be used for clinical or public health decision-making or presented without highlighting these facts. For more information, please see the FAQs.
Interpretable Artificial Intelligence in Assisting Treatment Response Prediction for Locally Advanced Rectal Cancer: A Prospective, Multicenter, Human-Model Interaction Study
25 Pages Posted: 2 Jul 2024
More...Abstract
Background: Preoperative assessment of pathological complete response (pCR) to neoadjuvant therapy (NAT) is an urgent need for anorectal preservation in patients with locally advanced rectal cancer (LARC). Artificial intelligence assistance remains challenging for lacking prospective validation and reliable interpretability.
Methods: Eligible LARC patients were retrospectively collected. Radiomic features extracted from post-NAT magnetic resonance imaging were applied to train a Deep Residual Shrinkage Network (DRSN) to generate Radscore for pCR probability. DRSN was further integrated with significant clinicopathological factors to construct a multimodality model, named as RAPIDS-II, in training set. RAPIDS-II performance in pCR prediction was verified in a testing set and further confirmed in a multicenter, prospective validation trial (NCT04278274). The improvements of radiologists visual assessment with RAPIDS-II assistance were evaluated in this prospective cohort. Area under curve (AUC) was used as primary endpoint for model performance.
Findings: Retrospectively recruited 823 LARC patients were divided into training set (n=575) and testing set (n=248). Compared to DRSN model, RAPIDS-II showed a comparable AUC of 0.813 [95% confidence interval (95%CI) 0.736-0.874] in testing set (P=0.020). In the prospective validation cohort (n=207), RAPIDS-II performed robustly with AUC of 0.795 (95%CI 0.723-0.859) in identifying pCR patients. Importantly, RAPIDS-II assistance resulted in improvements in overall AUC and sensitivity of radiologists’ visual assessment, especially for junior radiologists. Interpretable SHapley Additive exPlanations analysis identified that Radscore attributed most to RAPIDS-II prediction.
Interpretation: Interpretable RAPIDS-II model performed well in pCR prediction, providing a reliable tool in real world scenarios to tailor therapy for LARC patients.
Trial Registration: ClinicalTrials.gov identifier NCT04278274.
Funding: National Science Fund for Distinguished Young Scholars (No.82225040); National Key Research and Development Program of China (No.2022YFC2503700, No.2022YFA1105300); Natural Science Foundation of Guangdong Province (No. 2023A1515011414); National Science Fund for Outstanding Young Scholars (No.82122057); Natural Science Foundation of China (No.82171163, No.82302307); and National Postdoctoral Program for Innovative Talents of China (No.BX20220359).
Declaration of Interest: All authors declare no competing interests.
Ethical Approval: The study was conducted in accordance with Declaration of Helsinki and approved by institutional ethic committees. The study protocol was designed and reported according to the TRIPOD statement specific to machine learning, and SPIRIT-AI and CONSORT-AI extension guidelines, respectively.
Keywords: Artificial Intelligence, Radiomics, Interpretable AI-Human Interaction, Machine Learning, Pathological Complete Response, Rectal Cancer
Suggested Citation: Suggested Citation