How many slices are required for accurate tumour pathological response assessment? A computational modelling approach on real-world solid tumours

<title>Abstract</title> <p>Background and Objective Pathological response assessment, the microscopic quantification of the percentage residual viable tumour removed at surgery after neoadjuvant therapy, is a key measure in clinical cancer management. Standardised methods based on expert consensus have been developed. Thorough tumour tissue sampling is recommended to accurately assign the extent of tumour response, but this potentially creates substantial workloads for laboratories and pathologists. In this work, we sought to use mathematical modelling to investigate optimal sampling strategies for pathological response assessment. Methods To investigate optimal strategies, we conducted 966 simulations on 322 real-world kidney tumour volumes extracted from CT images, repeated in four virtual populations characterised by different frequencies of cancer treatment response. We calculated accuracy of pathological response classification for varying numbers of evenly spaced tumour slices. Key Findings and Limitations Across all four populations, five evenly spaced slices per tumour achieved &gt; 95% accuracy for classifying pathological complete response, and ~ 90% accuracy for major response, partial response and non-response. For continuous classification of residual viable tumour to the nearest 10%, more slices were needed—at least 10 for ~ 90% accuracy. Conclusions and Clinical Implications These models suggest a fixed number of tissue slices per tumour may be adequate to identify pathological complete response with high confidence. These simulations may form the basis for real-world examination of the performance of these strategies which, if validated, may reduce the workload on pathology services.</p>