Generalization optimizing machine learning to improve CT scan radiomics and assess immune checkpoint inhibitors' response in non-small cell lung cancer: a multicenter cohort study

Background

Recent developments in artificial intelligence suggest that radiomics may represent a promising non-invasive biomarker to predict response to immune checkpoint inhibitors (ICIs). Nevertheless, validation of radiomics algorithms in independent cohorts remains a challenge due to variations in image acquisition and reconstruction. Using radiomics, we investigated the importance of scan normalization as part of a broader machine learning framework to enable model external generalizability to predict ICI response in non-small cell lung cancer (NSCLC) patients across different centers.

Methods

Radiomics features were extracted and compared from 642 advanced NSCLC patients on pre-ICI scans using established open-source PyRadiomics and a proprietary DeepRadiomics deep learning technology. The population was separated into two groups: a discovery cohort of 512 NSCLC patients from three academic centers and a validation cohort that included 130 NSCLC patients from a fourth center. We harmonized images to account for variations in reconstruction kernel, slice thicknesses, and device manufacturers. Multivariable models, evaluated using cross-validation, were used to estimate the predictive value of clinical variables, PD-L1 expression, and PyRadiomics or DeepRadiomics for progression-free survival at 6 months (PFS-6).

Results

The best prognostic factor for PFS-6, excluding radiomics features, was obtained with the combination of Clinical + PD-L1 expression (AUC = 0.66 in the discovery and 0.62 in the validation cohort). Without image harmonization, combining Clinical + PyRadiomics or DeepRadiomics delivered an AUC = 0.69 and 0.69, respectively, in the discovery cohort, but dropped to 0.57 and 0.52, in the validation cohort. This lack of generalizability was consistent with observations in principal component analysis clustered by CT scan parameters. Subsequently, image harmonization eliminated these clusters. The combination of Clinical + DeepRadiomics reached an AUC = 0.67 and 0.63 in the discovery and validation cohort, respectively. Conversely, the combination of Clinical + PyRadiomics failed generalizability validations, with AUC = 0.66 and 0.59.

Conclusion

We demonstrated that a risk prediction model combining Clinical + DeepRadiomics was generalizable following CT scan harmonization and machine learning generalization methods. These results had similar performances to routine oncology practice using Clinical + PD-L1. This study supports the strong potential of radiomics as a future non-invasive strategy to predict ICI response in advanced NSCLC.

Toxicity of metal oxide nanoparticles in Escherichia coli correlates with conduction band and hydration energies

Metal oxide nanoparticles (MOx NPs) are used for a host of applications, such as electronics, cosmetics, construction, and medicine, and as a result, the safety of these materials to humans and the environment is of considerable interest. A prior study of 24 MOx NPs in mammalian cells revealed that some of these materials show hazard potential. Here, we report the growth inhibitory effects of the same series of MOx NPs in the bacterium Escherichia coli and show that toxicity trends observed in E. coli parallel those seen previously in mammalian cells. Of the 24 materials studied, only ZnO, CuO, CoO, Mn2O3, Co3O4, Ni2O3, and Cr2O3 were found to exert significant growth inhibitory effects; these effects were found to relate to membrane damage and oxidative stress responses in minimal trophic media. A correlation of the toxicological data with physicochemical parameters of MOx NPs revealed that the probability of a MOx NP being toxic increases as the hydration enthalpy becomes less negative and as the conduction band energy approaches those of biological molecules. These observations are consistent with prior results observed in mammalian cells, revealing that mechanisms of toxicity of MOx NPs are consistent across two very different taxa. These results suggest that studying nanotoxicity in E. coli may help to predict toxicity patterns in higher organisms.

Relating Nanoparticle Properties to Biological Outcomes in Exposure Escalation Experiments

A fundamental goal in nano-toxicology is that of identifying particle physical and chemical properties, which are likely to explain biological hazard. The first line of screening for potentially adverse outcomes often consists of exposure escalation experiments, involving the exposure of micro-organisms or cell lines to a library of nanomaterials. We discuss a modeling strategy, that relates the outcome of an exposure escalation experiment to nanoparticle properties. Our approach makes use of a hierarchical decision process, where we jointly identify particles that initiate adverse biological outcomes and explain the probability of this event in terms of the particle physicochemical descriptors. The proposed inferential framework results in summaries that are easily interpretable as simple probability statements. We present the application of the proposed method to a data set on 24 metal oxides nanoparticles, characterized in relation to their electrical, crystal and dissolution properties.

Zebrafish high-throughput screening to study the impact of dissolvable metal oxide nanoparticles on the hatching enzyme, ZHE1

The zebrafish is emerging as a model organism for the safety assessment and hazard ranking of engineered nanomaterials. In this Communication, the implementation of a roboticized high-throughput screening (HTS) platform with automated image analysis is demonstrated to assess the impact of dissolvable oxide nanoparticles on embryo hatching. It is further demonstrated that this hatching interference is mechanistically linked to an effect on the metalloprotease, ZHE 1, which is responsible for degradation of the chorionic membrane. The data indicate that 4 of 24 metal oxide nanoparticles (CuO, ZnO, Cr2 O3 , and NiO) could interfere with embryo hatching by a chelator-sensitive mechanism that involves ligation of critical histidines in the ZHE1 center by the shed metal ions. A recombinant ZHE1 enzymatic assay is established to demonstrate that the dialysates from the same materials responsible for hatching interference also inhibit ZHE1 activity in a dose-dependent fashion. A peptide-based BLAST search identifies several additional aquatic species that express enzymes with homologous histidine-based catalytic centers, suggesting that the ZHE1 mechanistic paradigm could be used to predict the toxicity of a large number of oxide nanoparticles that pose a hazard to aquatic species.

Use of metal oxide nanoparticle band gap to develop a predictive paradigm for oxidative stress and acute pulmonary inflammation

We demonstrate for 24 metal oxide (MOx) nanoparticles that it is possible to use conduction band energy levels to delineate their toxicological potential at cellular and whole animal levels. Among the materials, the overlap of conduction band energy (E(c)) levels with the cellular redox potential (-4.12 to -4.84 eV) was strongly correlated to the ability of Co(3)O(4), Cr(2)O(3), Ni(2)O(3), Mn(2)O(3), and CoO nanoparticles to induce oxygen radicals, oxidative stress, and inflammation. This outcome is premised on permissible electron transfers from the biological redox couples that maintain the cellular redox equilibrium to the conduction band of the semiconductor particles. Both single-parameter cytotoxic as well as multi-parameter oxidative stress assays in cells showed excellent correlation to the generation of acute neutrophilic inflammation and cytokine responses in the lungs of C57 BL/6 mice. Co(3)O(4), Ni(2)O(3), Mn(2)O(3), and CoO nanoparticles could also oxidize cytochrome c as a representative redox couple involved in redox homeostasis. While CuO and ZnO generated oxidative stress and acute pulmonary inflammation that is not predicted by E(c) levels, the adverse biological effects of these materials could be explained by their solubility, as demonstrated by ICP-MS analysis. These results demonstrate that it is possible to predict the toxicity of a large series of MOx nanoparticles in the lung premised on semiconductor properties and an integrated in vitro/in vivo hazard ranking model premised on oxidative stress. This establishes a robust platform for modeling of MOx structure-activity relationships based on band gap energy levels and particle dissolution. This predictive toxicological paradigm is also of considerable importance for regulatory decision-making about this important class of engineered nanomaterials.