Mitchell Chen

Background: Survival prediction in early‐phase trial participants is crucial for ensuring only those with sufficient life expectancy are offered experimental therapies. This study evaluated automated CT‐derived sarcopenia metrics for predicting overall survival (OS) and clinically significant treatment‐related adverse events (CS‐AE) in early‐phase immunotherapy trials. Methods: Participants with cancers of various primaries from early‐phase immunotherapy trials at two centres were included (Imperial College London, United Kingdom: n = 53, male:female (M:F) = 23:30, age (mean ± standard deviation): 59.4 ± 15.4; Humanitas Cancer Centre, Milan, Italy: n = 58, M:F = 32:26, age (mean ± standard deviation): 64.8 ± 11.6). The most common primary was ovarian (18.9%) in London and hepatocellular (31.0%) in Milan . 58.5% and 32.8% received single‐agent therapy in London and Milan , respectively and the remainder combination therapy. We computed sarcopenia metrics, including skeletal muscle area (SMA) and mean Hounsfield unit (MHU), using in‐house AI software, SarcAI , imputed a sex‐specific composite sarcopenia score (CSS) using Cox regression, based on height‐adjusted SMA, MHU and body mass index (BMI), and benchmarked their predictive performance for OS and CS‐AE. Results: Overall, 66 prognostic events were observed, with a 1‐year mortality rate of 86.8% (95% confidence interval [CI]: 73.7%–93.4%) in London and 35.7% (95% CI: 21.9%–47.1%) in Milan . SMA predicted 1‐year OS in female patients (area under the receiver operating characteristic curve 0.75, 95% CI [0.51, 0.98]). MHU predicted 90‐day OS (0.76 [0.54, 0.98]) and CSS 1‐year OS (0.70 [0.53, 0.87]). Both exceeded the prognostic performance of performance status (PS) in Milan (1‐year: 0.61 [0.43, 0.80]; 90‐day: 0.56 [0.13, 0.98]). SMA and CSS showed borderline significant prediction for CS‐AE (SMA: 0.64 [0.44, 0.84]; CSS: 0.66 [0.47, 0.86]), outperforming PS (0.59 [0.41, 0.77]). Conclusions: AI‐based CT sarcopenia metrics showed associations with OS and CS‐AE in early‐phase immunotherapy trial participants, supporting further investigation into their clinical predictive utility.

Background: Cardiovascular disease (CVD) and non-small cell lung cancer (NSCLC) are the global leading causes of overall and cancer-related deaths, respectively. NSCLC patients have a higher CVD risk than the general population which is frequently underdiagnosed. Coronary artery calcification (CAC), a marker of CVD, is commonly detected on routinely acquired CT from NSCLC work-up but often not reported. We present an automated CAC assessment tool validated for NSCLC patients using a deep learning-based framework to provide a non-invasive CVD screening opportunity without incurring extra workload or radiation exposure. Methods: We trained nnU-Net models on ungated, unenhanced chest CTs ( n = 97) from Stanford AIMI dataset, and tested them on three mutually independent datasets: (1) ungated unenhanced CTs from AIMI ( n = 95), (2) attenuation correction CTs from PET-CT scans of NSCLC patients at our institution (ICHNT, n = 87; age 67.8 ± 10.1 years; M:F 174:113), and (3) CAC-negative scans from TCIA ( n = 50); and used the best performing model to produce CAC segmentations, post-processed with TotalSegmentator, to stratify patients into CVD risk groups, informing the need for dedicated cardiac clinic assessment. Results: For a CAC threshold of 100, the model achieved accuracy: 83.6%, sensitivity: 91.9%, specificity: 70.8%, positive predictive value (PPV): 82.9%, negative predictive value (NPV): 85.1%, F1-score: 0.87, kappa coefficient: 0.65 and Area Under the Receiver Operating Characteristic Curve (AUC) score of 0.899. For a threshold of 400, accuracy: 84.5%, sensitivity: 90.9%, specificity: 79.5%, PPV: 77.6%, NPV: 91.8%, F1-score: 0.84, and kappa coefficient: 0.69 as well as an AUC of 0.926. Conclusion: Our optimised deep learning model can benefit NSCLC patients by providing CVD risk information from their routine CT scans which may not acted upon otherwise, thus enabling a practical opportunistic screening solution for these patients.

Purpose: Accurate prognostication is crucial for guiding personalised treatment strategies in non-small cell lung cancer (NSCLC). While radiomics offers promise, few features are derived from cancer models with causal justification to support their biological validity. This study evaluated the prognostic utility of normalised hotspot-to-centroid distance (NHOC), a recently proposed [18F]FDG PET imaging metric derived from a cancer evolutionary model, and its integration with PET/CT radiomics and clinical features to form a composite signature, non-invasive lung cancer evolution vector (nLCEV). Methods: A retrospective, multi-centre study was conducted using pre-treatment [18F]FDG PET/CT scans from 285 NSCLC patients (mean age: 67.7 ± 10.1 years; male:female = 171:114, International Association for the Study of Lung Cancer stage: T1/2/3/4/unknown = 61/118/53/52/1, N0/1/2/3/unknown = 133/46/71/34/1, M0/1/unknown = 222/62/1) from Imperial College Healthcare NHS Trust as the discovery cohort. External validation cohorts included patients from King’s College ( n = 53), Royal Marsden ( n = 63), Mount Vernon ( n = 61), and Nottingham University ( n = 38) hospitals. NHOC was evaluated for 3-year overall survival prediction and combined with a multi-regional PET/CT radiomics predictive vector (RPV) and disease stage to develop nLCEV. Results: NHOC and RPV demonstrated independent prognostic value (hazard ratio (HR) [95% confidence interval]: 2.52 [1.60–3.98] and 2.68 [2.13–3.38], respectively). nLCEV achieved an area under the receiver operating characteristic curve of 0.76 [0.60–0.92] and stratified patients into high- and low-risk groups across all validation cohorts with significant HR: KCL 3.27 [1.31, 8.16], Marsden 2.21 [1.02, 4.78], Mount Vernon 2.60 [1.42, 4.76], and Nottingham 4.14 [1.44, 11.90] (all p < 0.05). Conclusion: NHOC enhances NSCLC patient survival prediction, and when integrated with PET-CT radiomics and disease stage, offers a robust, non-invasive approach to disease prognostication.

Background: Newer generation tyrosine kinase inhibitors (TKI) are becoming more effective against cancers exhibiting certain driver mutations⁠, with those relating to epidermal growth factor receptor (EGFR) being the mostly commonly targeted in non-small cell lung cancer (NSCLC). Treatment decision is currently guided by tissue sampling and genetic testing⁠, an invasive process limited by patient tolerance, procedural complications, tumour heterogeneity and mutation evolution. Co-mutational status of EGFR with other target mutations can affect treatment response and is associated with worse patient outcomes. Exclusive EGFR mutation, where other common actionable mutations are negative, forms a favourable treatment scenario supporting first line EGFR-TKI use. Methods: We developed a radiomics signature, EGFR-RPV, to predict for exclusive EGFR mutational status, based on NSCLC patients (n = 304, age: 67.6 ± 10.7, male: female [M:F] = 174:130) who underwent CT scans and tissue sampling with genetic testing at our multi-centre institution between February 2012 and July 2018, and tested on an independent testing cohort of 51 patients (Age: 69.6 ± 8.1, M: F = 35:16); and conducted enrichment analysis of their paired RNA transcriptomics readouts. Results: EGFR-RPV predicted exclusive EGFR mutation to an accuracy of 0.77, 95% CI: 0.66–0.88 and 0.71, 95% CI: 0.54-0.89 in the internal and external testing sets, respectively. EGFR-RPV also achieved patient prognostic stratification (hazard ratio 2.15, 95% CI 1.50-3.08). We discovered the enrichment of FAM190A and CBMO1 genes in exclusive EGFR positive cases, concordant with their known regulatory roles in cancer cell division and conversion of beta-carotene into vitamin A, respectively. Conclusion: EGFR-RPV enables non-invasive prediction of exclusive EGFR mutations in NSCLC, with a potential to guide first-line EGFR-TKI use