Multitask Radiomics-based Dynamic Nomogram for Joint Prediction of ALK Mutation and Ki-67 Expression in NSCLC

Multitask Radiomics-based Dynamic Nomogram for Joint Prediction of ALK Mutation and Ki-67 Expression in NSCLC | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Multitask Radiomics-based Dynamic Nomogram for Joint Prediction of ALK Mutation and Ki-67 Expression in NSCLC Ting Lin, Shengwang Peng, Chuanyan Li, Yue Yu, Muyao Zhou, Xianyue Quan, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6829111/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 5 You are reading this latest preprint version Abstract Objectives Anaplastic lymphoma kinase (ALK) mutation and Ki-67 expression are clinically significant biomarkers for non-small-cell lung cancer (NSCLC) assessment and treatment. Most prior studies have focused on single biomarkers while a comprehensive consideration of multiple biomarkers is necessary in clinics. The study aimed to develop a multitask radiomics-based dynamic nomogram to simultaneously predict ALK mutational status and Ki-67 expression. Methods A total of 534 patients with pathologically confirmed NSCLC (via surgical resection or needle biopsy) were retrospectively enrolled. Multitask radiomics models were developed using preoperative CT imaging features to jointly predict ALK mutation status and Ki-67 expression levels. Model performance was evaluated through area under the receiver operating characteristic curve (AUC), net reclassification improvement (NRI), and decision curve analysis (DCA). A web-based interactive platform was implemented to visualize the dynamic nomogram. Principal component analysis (PCA) was applied to identify the most influential radiomic features driving model predictions. Results The multitask radiomics model significantly outperformed the clinical model in validation cohorts. For ALK mutation prediction, the AUC was 0.9974 (95%CI: 0.9923, 1.0000) in the training cohort and 0.9918 (95%CI: 0.9744-1.0000) in the validation cohort. For Ki-67 expression, the AUC was 0.9778 (95%CI: 0.9648–0.9878) for training and 0.9807 (95%CI: 0.9602–0.9950) for testing, respectively. NRI analysis confirmed significant reclassification improvements over traditional clinical model (p < 0.05), and DCA revealed a higher net clinical benefit across clinically relevant threshold probabilities. PCA leaded to significant improvements in both training cohort and test cohort. The feature contribution analysis identified key radiomic features, such as original_glszm_zoneentropy (capturing zone size heterogeneity in original images), as dominant contributors to model predictions. Conclusion The multitask radiomics-based dynamic nomogram demonstrated robust predictive performance for ALK mutation and Ki-67 expression, offering a noninvasive preoperative tool to guide personalized treatment in NSCLC. Radiomics anaplastic lymphoma kinase Ki-67 Non–small-cell lung cancer Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Key Points T his study developed a multitask radiomics nomogram using preoperative CT images to simultaneously predict ALK mutation status and Ki-67 expression in NSCLC patients, achieving exceptional accuracy (AUC >0.96) and overcoming limitations of traditional single-task models. The model outperformed clinical predictors, demonstrating clinical utility through improved reclassification and decision curve analysis, offering a noninvasive alternative to tissue biopsy. A web-based interactive platform enables real-time prediction updates, while PCA-revealed feature hierarchies highlight biologically plausible radiomic drivers, bridging the gap between artificial model and clinical interpretability. Introduction Non-small-cell lung cancer (NSCLC) is one of the most frequent cancer types and is responsible for most cancer deaths worldwide [ 1 ] . Despite advances in therapy, many NSCLC patients experience relapses [ 2 ] , highlighting the need for personalized treatment strategies [ 3 ] . Biomarkers such as Anaplastic lymphoma kinase (ALK) mutation and Ki-67 expression play pivotal roles in prognosis and clinical decision-making in NSCLC [ 4 , 5 ] . ALK rearrangement occur in 4–5% of NSCLC patients. Up to 50–60% of ALK-positive patients are at high risk for brain metastases [ 6 ] . Targeted therapies have shown efficacy in the treatment of oncogene-dependent NSCLC, which has the potential to improve outcomes [ 7 ] . Meanwhile, Ki-67, a marker of cellular proliferation, correlates with aggressive tumor behavior and poor prognosis [ 8 ] . Currently, these biomarkers are assessed via invasive tissue biopsies, which pose challenges such as sampling bias, procedural risks, and delayed results. Given the clinical importance of both biomarkers, a simultaneous prediction method could provide comprehensive insights for treatment planning. Computed tomography (CT) is the primary imaging modality for NSCLC evaluation. Radiomics, a high-throughput feature extraction technique, has shown promise in noninvasively characterizing tumor heterogeneity. While prior studies have focused on single biomarkers [ 9 , 10 ] , a comprehensive consideration of multiple biomarkers is necessary in clinics. It has been challenging to combine ALK mutational status and Ki-67 expression prediction, which were traditionally treated as separate problems in medical image analysis. A multitask approach could enhance predictive accuracy by leveraging shared radiomics features between ALK and Ki-67. In this study, we proposed a multitask radiomics model to simultaneously predict ALK mutation and Ki-67 expression. By integrating clinical and radiomics data, our dynamic nomogram aims to provide a practical, noninvasive tool for preoperative biomarker assessment. Materials and Methods Study Participants Ethical approval of this study was obtained from the Ethics Committees of Zhujiang Hospital and conducted in accordance with the ethical principles for medical research involving human subjects as described in the 1964 Helsinki Declaration and its later amendments. The requirement for informed consent was waived due to the retrospective nature of this study. This retrospective study included 534 NSCLC patients between January 2015 and December 2020 who met the following inclusion and exclusion criteria were enrolled in this study. Inclusion criteria were as follows: (a) pathologically confirmed NSCLC (surgical resection or biopsy); (b) preoperative CT images were available. Exclusion criteria were as follows: (a) prior chemotherapy/radiotherapy (n = 17); (b) multiple primary carcinomas or concurrent malignancies (n = 23); (c) poor quality of CT images (n = 34); (d) incomplete clinical data (n = 15). We randomly divided the patients into a training cohort and a test cohort through the 7:3 division. The statistical analysis of clinical characteristics between the two groups was shown in Table 1 . Table 1 Clinical characteristics of patients in training and test cohort Training cohort (n = 373) Test cohort (161) P value Age (years) 60.98 (14 ~ 88) 60.11 (27 ~ 85) 0.3754 Gender 0.2545 Male 234 (62.73%) 110 (68.32%) Female 139 (37.27%) 51 (31.68%) Smoking status 0.5022 No 242 (64.88%) 110 (68.32%) Yes 131 (35.12%) 51 (31.68%) Diameter 36.69 (3.44 ~ 144.90) 38.21 (10.00 ~ 139.50) 0.4665 Location 0.6865 LUL 108 (28.95%) 53 (32.92%) LLL 51 (13.67%) 24 (14.91%) RUL 116 (31.10%) 108(32.2) RML 25 (6.70%) 51 (31.68%) RLL 73 (19.57%) 25 (15.53%) Pathological type 0.6005 SCC 50 (13.40%) 25 (15.53%) ADC 297 (79.62%) 122 (75.78%) Other 26 (6.97%) 14 (8.70%) ALK 0.9186 Negative 337 (90.35%) 145 (90.06%) Positive 36 (9.65%) 16 (9.94%) Ki67 expression 0.4454 Low 254 (68.10%) 115 (71.43%) High 119 (31.90%) 46 (28.57%) Values are n (%) or mean (SD). P* were calculated by independent samples T test for Age/Diameter and Chi-square test for Gender/Smoking status/Location/Pathological type. RUL, right upper lobe; RML, right middle lobe; RLL, right lower lobe; LUL, left upper lobe; LLL, left lower lobe; GGO, ground-glass opacity; SCC, squamous cell carcinoma; ADC, adenocarcinoma. Clinical and Pathological Data Baseline clinical information was retrieved from the institutional database for medical records, including lesion density, smoking status, the maximum tumor diameter, tumor location, age, gender, and pathological type. ALK status was determined via immunohistochemistry (IHC, Ventana Benchmark XT), with positivity defined as strong cytoplasmic staining. Ki-67 expression was assessed using the CONFIRM anti-Ki-67 antibody, with high expression defined as > 30% nuclear staining. Image Acquisition and Segmentation CT scans were performed using Philips Brilliance iCT 256 or 64-slice scanners (120 kVp; field of view: 350mm×350mm; matrix, 512×512; slice thickness: 5mm; slice thickness: 5 mm). All CT images were obtained from the picture archiving and communication system (PACS). Tumor segmentation was manually performed by two radiologists (8- and 5-year experience in chest CT interpretation) using ITK-SNAP(version 3.8.0; http://www.itksnap.org/ ). Radiomics Feature Extraction and Model Development A total of 1437 IBSI-standardized radiomic features were extracted using the Standardized Environment for Radiomics Analysis (SERA). including: 771 first-order features (e.g. morphology, statistical, histogram and intensity-histogram features), 666 higher-order features (Supplementary table1). Besides, 6 clinical features are included (age, gender, smoke, pathological type, location, diameter). After standardizing the features using Z-score normalization to ensure uniform scale across all dimensions, variance thresholding was applied as a basic feature selection method to remove low-variance features (threshold = 0.01), aiming to eliminate redundant or nearly constant features. Subsequently, Principal Component Analysis (PCA) was performed to further reduce dimensionality by transforming the data into a lower-dimensional space that retains 95% of the original variance, with additional analysis conducted to examine the explained variance ratio per principal component and the most influential original features contributing to each component. In the end, the feature dimension was reduced to 63. After that, we built the multi-task learning framework for the ALK and Ki-67 classification, where a base classifier \(\:{\mathcal{M}}_{\theta\:}\) is extended to handle multivariate target predictions through the MultiOutputClassifier formalism. Assuming there are t tasks (t = 2), features \(\:X\in\:{\mathcal{R}}^{N*D}(\text{D}=63)\) , and labels \(\:Y\in\:{\mathcal{R}}^{N*K}(K=2)\) , the composite model \(\:{\mathcal{F}}_{\varTheta\:}\) is formulated as: $$\:{\mathcal{F}}_{\varTheta\:}\left(X\right)\:=\:\left[\sigma\:\right({\mathcal{M}}_{{\theta\:}_{1}}\left(X\right)),\:\sigma\:({\mathcal{M}}_{{\theta\:}_{2}}\left(X\right))\:,...,\sigma\:({\mathcal{M}}_{{\theta\:}_{t}}\left(X\right)\left)\:\right]$$ Where, \(\:\sigma\:\) denotes the task decision model. Figure 1 depicts the workflow of the proposed model. Evaluation of Model Performance The receiver operating characteristic curves (ROCs) of the combined model and clinical model in both the training and validation groups were plotted, and the diagnostic accuracy, sensitivity, and specificity were calculated to evaluate the discriminative performance of the models. Delong ’s test was used to compare the ROC curves of models. The performance improvement introduced by the inclusion of multitask radiomics signature was quantified by net reclassification improvement (NRI). Furthermore, a decision curve analysis (DCA) was used to estimate clinical usefulness. Statistical Analysis We performed a comparison for each clinical information between the training group and validation group for NSCLC patients. Categorical variables were analyzed by either Chi-square test. Continuous variables were compared between groups using a student’s t-test for independent samples. All analyses were performed using Python 3.8 platform. All tests for statistical significance were two-tailed and P values less than 0.05 were considered statistically significant. Results Clinical Characteristics All patients were randomly allocated to the training cohort and validation cohort at a ratio of 7:3. Pathology confirmed 36 cases of ALK-positive (9.65%) in training cohort and 16 cases (9.94%) in validation cohort (Table 1 ). 119 cases showed high Ki-67 expression (31.90%) in training cohort and 46 cases (28.57%) in validation cohort (Table 1 ). Model Performance in Discriminating ALK Mutation and Ki-67 Expression The multitask integrative model showed better discrimination accuracy than clinical model (Table 2 , 3 ). The multitask integrative model yielded the higher AUC (0.9918 [95%CI: 0.9744-1.0000]) in predicting the ALK status compared with clinical model (0.5434 [95%CI: 0.3424–0.7249]). Meanwhile, The AUC of multitask integrative model for predicting the Ki67 expression was 0.9807 [95%CI: 0.9602–0.9950], which was superior to the clinical model (0.7044 [95%CI: 0.6229–0.7814]). Table 2 The predictive performance of the multi-task model and clinical model for ALK mutation in training cohort and test cohort Training cohort Test cohort Multi-task model Single-task model P value Multi-task model Clinical model P value AUC 0.9974 [95%CI: 0.9923, 1.0000] 0.9993 [95%CI: 0.9973-1.0000] 0.18 0.9918 [95%CI: 0.9744-1.0000] 0.5434 [95%CI: 0.3424–0.7249] < 0.05 Accuracy 0.9866 [95%CI: 0.9759–0.9973] 0.9783 [95%CI: 0.9625–0.9920] 0.9689 [95%CI: 0.9379–0.9938] 0.6282 [95%CI: 0.5528–0.7019] Sensitivity 0.9720 [95%CI: 0.9062-1.0000] 1.0000 [95%CI: 1.0000–1.0000] 0.9386 [95%CI: 0.7857-1.0000] 0.5691 [95%CI: 0.2857–0.8182] Specificity 0.9882 [95%CI: 0.9763–0.9971] 0.9760 [95%CI: 0.9592–0.9910] 0.9722 [95%CI: 0.9437–0.9932] 0.6346 [95%CI: 0.5555–0.7114] Table 3 The predictive performance of the multi-task model and clinical model for Ki-67 expression in training cohort and test cohort Training cohort Test cohort Multi-task model Clinical model P value Multi-task model Clinical model P value AUC 0.9778 [95%CI: 0.9648–0.9878] 0.9899 [95%CI: 0.9820–0.9957] < 0.05 0.9807 [95%CI: 0.9602–0.9950] 0.7044 [95%CI: 0.6229–0.7814] < 0.05 Accuracy 0.9033 [95%CI: 0.8740–0.9303] 0.9409 [95%CI: 0.9142–0.9651] 0.9187 [95%CI: 0.8758–0.9565] 0.6588 [95%CI: 0.5839–0.7267] Sensitivity 0.8907 [95%CI: 0.8349–0.9440] 0.9490 [95%CI: 0.9065–0.9835] 0.9119 [95%CI: 0.8205–0.9796] 0.7401 [95%CI: 0.6052–0.8628] Specificity 0.9091 [95%CI: 0.8740–0.9425] 0.9371 [95%CI: 0.9057–0.9656] 0.9213 [95%CI: 0.8700-0.9658] 0.6263 [95%CI: 0.5366–0.7155] Feature Preprocessing Strategy Analysis The ROC curves of the multitask integrative model to predict the ALK status and Ki67 expression using original features and processed features are shown in Fig. 2 . After computing the prior components using PCA analysis, model performance obtained a significant improvement in both training cohort and test cohort. It may be due to the ability in avoiding the negative effects of some noisy features. Importantly, the gains were consistently large in both data groups, indicating that the performance gains were mostly due to the powerful feature representations which lead to better convergence. Net Reclassification Improvement Analysis of Our method and Clinical Model The Net Reclassification Improvement (NRI) results in Table 4 demonstrate that our method achieves significant performance gains over clinical model particularly on the test cohort. For the ALK biomarker, the test set NRI of 0.727 indicates a substantial improvement in classification accuracy, with 0.438 (NRI+) reflecting enhanced classification ability of positive cases and 0.289 (NRI-) showing improved downward reclassification of negative cases. It means that the performance gains majorly come from the correction of positive cases. Similarly, the Ki-67 biomarker exhibits a test NRI of 0.387, driven by complementary gains in both NRI+ (0.109) and NRI- (0.278). Rather, the performance gains of Ki-67 expression is due to the correction of negative cases. In contrast, the training data shows mixed results, with ALK (NRI = -0.040) and Ki67 (NRI = -0.139) displaying minor degradation in reclassification performance. This discrepancy between training and test metrics may arise from regularization effects or overfitting mitigation in our framework, which prioritizes generalization over training data alignment. These findings underscore the value of our approach in enhancing ALK and Ki-67 prediction reliability at the same time, particularly for such imbalanced classification tasks. Decision Curve Analysis of Our Prediction Model The Decision Curve Analysis (DCA) in Fig. 3 demonstrates that our model provides significant clinical utility for both ALK and Ki-67 predictions. For ALK, the model outperforms "treat all" and "treat none" strategies across most threshold probabilities (0.2–0.8), with peak net benefit at intermediate thresholds (0.3–0.6), indicating optimal decision-making for balancing over-treatment avoidance and sensitivity in clinical practice. For Ki-67, the model shows superior net benefit at low thresholds (< 0.4), highlighting its value in early risk identification for high-risk patients, while maintaining utility even at higher thresholds. These results suggest that integrating our model into clinical workflows could improve personalized risk stratification, reduce unnecessary interventions. Operating System of Our Multi-task Prediction Model We built an operating interface for dynamic nomograms predicting ALK status and Ki67 expression in patients as shown in Fig. 4 . It comprises the input groups on the left and the output groups on the right. Operators can upload a participant’s CT Images and the tumor delineations at first and the system will automatically compute the radiomics fast. Then, operators could enter the clinical features, i.e., age, gender, pathological type, tumor size and location in this interface. The corresponding probability and 95% confidence intervals of ALK status and Ki67 expression will be shown in the right. PCA-Revealed Feature Contribution Analysis The heatmap analysis (Fig. 5 ) of PCA-derived feature contributions reveals a hierarchical structure in feature importance, quantifying the top 20 features' loadings on the first 20 principal components (PCs). A subset of radiomic features, including original_glm_ZoneEntropy and logarithm_ngth_Correlation, dominates variance explanation in early PCs (PC1–PC5), which primarily capture global textural descriptors. Distinct contribution patterns are observed: features such as wavelet-LH1_LLL_glm_ClusterProminence exhibit consistent high positive or negative loadings (± 0.05–±0.10) across multiple PCs, indicating their robustness in capturing orthogonal data variations. Conversely, features like wavelet-LLL_glcm_ClusterShade show negligible contributions ( < ± 0.02), suggesting redundancy in the reduced-dimensional space. Later PCs (PC10–PC20) reflect specialized variations driven by features such as lbp-3D-k_glrlm_RunLengthNoUniformityNormalized. This stratified feature contribution pattern demonstrates PCA's capacity to prioritize informative features for downstream modeling while suppressing noise. Discussion The identification of biomarkers is important for NSCLC patients to determine the suitability of treatment and therapeutic outcomes, which is critical in precision medicine for NSCLC patients [ 11 , 12 ] . Due to the limitations of invasive biopsy or surgery, constructing an easy-to-use model based on easily available noninvasive clinical information has become a concern. In this study, the multitask integrative models were built as noninvasive tools to determine the ALK mutation and Ki-67 expression status in patients with NSCLC. The present results suggested that it is feasible to use multitask machine learning based model to predict multiple clinical indicators simultaneously. Our results showed that ALK mutation was common in younger patients and male patients with adenocarcinoma are more prone to high Ki-67 expression. Our findings align with previous studies showing the predictive value of radiomics for ALK mutation, while extending the approach to include Ki-67 expression [ 13 , 14 ] . Although the above factors played an important role in the identification of ALK mutation and Ki-67 expression, the clinical model constructed with these factors had a relatively low AUC value. Compared with the clinical model, multitask integrative models improve the prediction of ALK mutation and Ki-67 expression simultaneously. Our study found that the multitask integrative model can predict ALK mutation in NSCLC, with an AUC of 0.9918 in validation cohort. Tan et al [ 14 ] built a model to distinguish EGFR mutant status and ALK rearrangement concurrently with an AUC of 0.70, possibly because of the mutually exclusive status between EGFR and ALK alterations. The assessment of ALK mutation and Ki-67 expression is important for prognosis and treatment management [ 15 , 16 ] . Ki-67 is a nuclear protein that is associated with cellular proliferation, conveying significant prognostic and predictive value in a variety of solid tumors [ 17 ] . ALK could stimulate cell proliferation and inhibit apoptosis. ALK mutated tumors display more aggressive behavior [ 18 ] . The inter-task correlation allows multitask machine learning exploits similar and complementary evaluation of tumor characteristics, hence improving the classification performance of model. Ganeshan et al [ 19 ] found that the texture features of NSCLC on CT were significantly correlated with the expression of tumor glucose transporter 1 (Glut-1) and tumor CD34, indicating that the CT radiomics features of the lesion reflected tumor angiogenesis and histopathological changes, such as hypoxia. However, those markers indirectly reflect tumor proliferation. In our study, machine learning aided radiomics approaches were applied to detect the Ki-67 expression, which can directly reflect tumor proliferation. Multitask algorithm has been applied in previous studies. Qi et al [ 20 ] provided a deep learning approach in lung cancer for automatically fast-tracking tumor lesions and classifying corresponding histological subtypes. Dong et al [ 21 ] proposed a multitasking deep learning model for predicting EGFR and KRAS mutation simultaneously and outperform single predictions. In our study, we selected radiomics features correlated with ALK mutation and Ki-67 expression based on multitask machine learning. The shared feature selection process may capture common tumor characteristics relevant to both biomarkers to improve robustness. Our method is a non-invasive auxiliary detection method suitable for avoiding invasive damage when surgery and biopsy are not convenient. Also, CT images are easily available throughout the treatment period to monitor ALK mutation and Ki-67 expression status. The acquisition of CT images is relatively inexpensive in terms of cost and time. Moreover, our method does not require physicians’ domain knowledge, facilitating the economical and convenient prediction of ALK mutation and Ki-67 expression. However, the present study has certain limitations. First, this retrospective study was conducted in one center. Ideally, a prospective multicenter study would enhance the conclusion of this study. Further research is necessary to test the generalizability of our models in other populations. Second, statistical modeling based on radiomics requires large samples to obtain optimized classifiers for prediction, but our sample size was not large enough. Third, more clinical characteristics should be included to strengthen the performance and outcomes of this study. Therefore, additional prospective data collection should be involved in future study. In conclusion, the proposed nomogram could serve as a preoperative decision-support tool, enabling personalized treatment planning without invasive procedures.” Abbreviations NSCLC Non-small cell lung cancer ALK Anaplastic lymphoma kinase TKIs Tyrosine kinase inhibitors IHC Immunohistochemistry ROI Region of interest ICC Interclass correlation coefficients SERA Standardized Environment for Radiomics Analysis IBSI Image Biomarker Standardization Initiative ROCs Receiver operating characteristic curves NRI Net reclassification improvement DCA Decision curve analysis Declarations Ethics approval and consent to participate Ethical approval of this study was obtained from the Ethics Committees of Zhujiang Hospital. The requirement for informed consent was waived due to the retrospective nature of this study. Consent for publication The requirement for informed consent was waived due to the retrospective nature of this study. Availability of data and materials The data that support the findings of this study are available on request from the corresponding author upon reasonable request. Competing interests None. Funding This work was financially supported by Science and Technology Program of Guangzhou (2024A04J5118), and President Foundation of Zhujiang Hospital, Southern Medical University (yzjj2022qn19, yzjj2022qn33). Authors’ contributions Ting Lin: Writing – original draft, Visualization, Methodology, Investigation, Funding acquisition. Shengwang Peng: Writing – original draft, Visualization, Methodology, Investigation. Chuanyan Li: Methodology, Investigation. Yue Yu: Methodology, Investigation. Muyao Zhou: Investigation. Xianyue Quan: Methodology, Investigation. Xin Chen: Writing – review & editing, Supervision. Jianbing Zhu: Writing – review & editing, Funding acquisition, Conceptualization. Dong Zeng: Writing – review & editing, Supervision, Resources, Conceptualization. Acknowledgments The authors acknowledge all contributors who assisted in this study. References Hendriks LEL, Remon J, Faivre-Finn C, Garassino MC, Heymach JV, Kerr KM, et al. Non-small-cell lung cancer. Nat reviews Disease primers. 2024;10(1):71. Chang JY, Lin SH, Dong W, Liao Z, Gandhi SJ, Gay CM, et al. Stereotactic ablative radiotherapy with or without immunotherapy for early-stage or isolated lung parenchymal recurrent node-negative non-small-cell lung cancer: an open-label, randomised, phase 2 trial. Lancet. 2023;402(10405):871–81. Wang M, Herbst RS, Boshoff C. Toward personalized treatment approaches for non-small-cell lung cancer. Nat Med. 2021;27(8):1345–56. Pike LRG, Miao E, Boe LA, Patil T, Imber BS, Myall NJ, et al. Tyrosine Kinase Inhibitors With and Without Up-Front Stereotactic Radiosurgery for Brain Metastases From EGFR and ALK Oncogene-Driven Non-Small Cell Lung Cancer (TURBO-NSCLC). J Clin Oncol. 2024;42(30):3606–17. Yang Y, Shao X, Li Z, Zhang L, Yang B, Jin B, et al. Prognostic heterogeneity of Ki67 in non-small cell lung cancer: A comprehensive reappraisal on immunohistochemistry and transcriptional data. J Cell Mol Med. 2024;28(14):e18521. Nagpal S, Milano MT, Chiang VL, Soltys SG, Brackett A, Halasz LM, et al. Executive summary of the American Radium Society appropriate use criteria for brain metastases in epidermal growth factor receptor mutated-mutated and ALK-fusion non-small cell lung cancer. Neurooncology. 2024;26(7):1195–212. Meyer ML, Fitzgerald BG, Paz-Ares L, Cappuzzo F, Jänne PA, Peters S, et al. New promises and challenges in the treatment of advanced non-small-cell lung cancer. Lancet. 2024;404(10454):803–22. Atari M, Imai K, Nanjo H, Wakamatsu Y, Takashima S, Kurihara N, et al. Rapid intraoperative Ki-67 immunohistochemistry for lung cancer using non-contact alternating current electric field mixing. Lung cancer (Amsterdam Netherlands). 2022;173:75–82. Prelaj A, Miskovic V, Zanitti M, Trovo F, Genova C, Viscardi G, et al. Artificial intelligence for predictive biomarker discovery in immuno-oncology: a systematic review. Annals oncology: official J Eur Soc Med Oncol. 2024;35(1):29–65. Chen M, Lu H, Copley SJ, Han Y, Logan A, Viola P, et al. A Novel Radiogenomics Biomarker for Predicting Treatment Response and Pneumotoxicity From Programmed Cell Death Protein or Ligand-1 Inhibition Immunotherapy in NSCLC. J Thorac Oncol. 2023;18(6):718–30. Riely GJ, Wood DE, Ettinger DS, Aisner DL, Akerley W, Bauman JR, et al. Non-Small Cell Lung Cancer, Version 4.2024, NCCN Clinical Practice Guidelines in Oncology. J Natl Compr Cancer Network: JNCCN. 2024;22(4):249–74. Tan AC, Tan DSW. Targeted Therapies for Lung Cancer Patients With Oncogenic Driver Molecular Alterations. J Clin Oncol. 2022;40(6):611–25. Gu Q, Feng Z, Liang Q, Li M, Deng J, Ma M, et al. Machine learning-based radiomics strategy for prediction of cell proliferation in non-small cell lung cancer. Eur J Radiol. 2019;118:32–7. Tan X, Li Y, Wang S, Xia H, Meng R, Xu J, et al. Predicting EGFR mutation, ALK rearrangement, and uncommon EGFR mutation in NSCLC patients by driverless artificial intelligence: a cohort study. Respir Res. 2022;23(1):132. Schneider JL, Lin JJ, Shaw AT. ALK-positive lung cancer: a moving target. Nat cancer. 2023;4(3):330–43. Wang J, Dong L, Zheng Z, Zhu Z, Xie B, Xie Y, et al. Effects of different KRAS mutants and Ki67 expression on diagnosis and prognosis in lung adenocarcinoma. Sci Rep. 2024;14(1):4085. Uxa S, Castillo-Binder P, Kohler R, Stangner K, Müller GA, Engeland K. Ki-67 gene expression. Cell Death Differ. 2021;28(12):3357–70. Cooper AJ, Sequist LV, Lin JJ. Third-generation EGFR and ALK inhibitors: mechanisms of resistance and management. Nat reviews Clin Oncol. 2022;19(8):499–514. Ganeshan B, Goh V, Mandeville HC, Ng QS, Hoskin PJ, Miles KA. Non-small cell lung cancer: histopathologic correlates for texture parameters at CT. Radiology. 2013;266(1):326–36. Qi J, Deng Z, Sun G, Qian S, Liu L, Xu B. One-step algorithm for fast-track localization and multi-category classification of histological subtypes in lung cancer. Eur J Radiol. 2022;154:110443. Dong Y, Hou L, Yang W, Han J, Wang J, Qiang Y, et al. Multi-channel multi-task deep learning for predicting EGFR and KRAS mutations of non-small cell lung cancer on CT images. Quant imaging Med Surg. 2021;11(6):2354–75. Supplementary Table Supplementary Table 1 is not available with this version. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviewers invited by journal 03 Jul, 2025 Editor invited by journal 09 Jun, 2025 Editor assigned by journal 08 Jun, 2025 Submission checks completed at journal 08 Jun, 2025 First submitted to journal 05 Jun, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6829111","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":480246391,"identity":"9facf502-80d4-49ed-bf16-72b88330340e","order_by":0,"name":"Ting Lin","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA3ElEQVRIiWNgGAWjYDCCA2DSgoeBgbGBgaFCQk6eCC0gpRI8DGwg+oyFsWEDkVoYGNiAHMa2ikSovbgB3/Hm5w8+7pGQMZdvbnv4dZ5EAmMD88NHN/BokTxzzLBxxjMJHss2xnZj2W0SeewMbMbGOXi0GNzIYWzmOSDBY3CMsU1acptEMWMDD5s0CVrmSCQ2HCBFi+THBiK0gPwycwZYS2KbNMMxCWPDZgJ+AYbYgw8fDtjYGxw+/kzyR02dnDx788PH+LSgAGYeMEmschBg/EGK6lEwCkbBKBgxAAAYHEsDqzvG8wAAAABJRU5ErkJggg==","orcid":"","institution":"Zhujiang Hospital","correspondingAuthor":true,"prefix":"","firstName":"Ting","middleName":"","lastName":"Lin","suffix":""},{"id":480246392,"identity":"8fff31b3-0301-4533-9ffe-c5e69b7b04ee","order_by":1,"name":"Shengwang Peng","email":"","orcid":"","institution":"Southern Medical University","correspondingAuthor":false,"prefix":"","firstName":"Shengwang","middleName":"","lastName":"Peng","suffix":""},{"id":480246393,"identity":"7aada0ae-8cbc-43b1-b888-31f6316eaaf3","order_by":2,"name":"Chuanyan Li","email":"","orcid":"","institution":"Shanghai Pudong New Area Gongli Hospital","correspondingAuthor":false,"prefix":"","firstName":"Chuanyan","middleName":"","lastName":"Li","suffix":""},{"id":480246394,"identity":"32c2d7a1-f1d4-449a-b084-5d6abe65c43a","order_by":3,"name":"Yue Yu","email":"","orcid":"","institution":"Zhujiang Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yue","middleName":"","lastName":"Yu","suffix":""},{"id":480246395,"identity":"38c3f7a2-9f59-4e32-bcac-7f7387f8dfad","order_by":4,"name":"Muyao Zhou","email":"","orcid":"","institution":"Guangdong Sanjiu Brain Hospital","correspondingAuthor":false,"prefix":"","firstName":"Muyao","middleName":"","lastName":"Zhou","suffix":""},{"id":480246396,"identity":"0612b1f9-a8bd-426e-aea2-d878f954e868","order_by":5,"name":"Xianyue Quan","email":"","orcid":"","institution":"Zhujiang Hospital","correspondingAuthor":false,"prefix":"","firstName":"Xianyue","middleName":"","lastName":"Quan","suffix":""},{"id":480246397,"identity":"0fd1eae7-0e45-4a8f-9f44-89db0f91b54d","order_by":6,"name":"Xin Chen","email":"","orcid":"","institution":"Guangzhou First People's Hospital","correspondingAuthor":false,"prefix":"","firstName":"Xin","middleName":"","lastName":"Chen","suffix":""},{"id":480246398,"identity":"9c9cd0e4-722e-41be-9ed2-a672c74ec824","order_by":7,"name":"Jianbing Zhu","email":"","orcid":"","institution":"Zhujiang Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jianbing","middleName":"","lastName":"Zhu","suffix":""},{"id":480246399,"identity":"4b1e8ba6-53fd-4b3f-9863-ce880befd5e8","order_by":8,"name":"Dong Zeng","email":"","orcid":"","institution":"Southern Medical University","correspondingAuthor":false,"prefix":"","firstName":"Dong","middleName":"","lastName":"Zeng","suffix":""}],"badges":[],"createdAt":"2025-06-05 12:23:28","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6829111/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6829111/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":86246677,"identity":"d6f6f9dc-a764-4388-b502-ed43dbe10dbd","added_by":"auto","created_at":"2025-07-08 11:45:56","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":113920,"visible":true,"origin":"","legend":"\u003cp\u003eWorkflow for developing multi-task radiomics models for predicting ALK mutation status and Ki-67 expression.\u003c/p\u003e","description":"","filename":"image1.png","url":"https://assets-eu.researchsquare.com/files/rs-6829111/v1/550adc4248014cd524d6b8dd.png"},{"id":86246673,"identity":"1e6af5bf-07ae-48fe-b731-484c770291fa","added_by":"auto","created_at":"2025-07-08 11:45:56","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":322028,"visible":true,"origin":"","legend":"\u003cp\u003eReceiver operating characteristic (ROC) curves of the multi-task machine learning model for predicting ALK status (A, B) and Ki67 expression (C, D) in train cohort (left) and test cohort (right). Red curves denote the model using features after PCA operation and blue curved denotes the model using original features.\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-6829111/v1/05f50911089deba4c7bca17a.png"},{"id":86246676,"identity":"75738a85-173e-4b4e-bd16-2fccd18db4e7","added_by":"auto","created_at":"2025-07-08 11:45:56","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":52191,"visible":true,"origin":"","legend":"\u003cp\u003eDecision curve analysis (DCA) of the multi-task integrative model predicting the ALK mutation (A) and Ki67 expression (B) in test cohort. The x-axis shows the risk threshold, the vertical axis shows the net benefit of standardization.\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-6829111/v1/8272e644f25e38538e4513d9.png"},{"id":86247685,"identity":"f0c39a94-0746-4412-9f32-da7bfdaac05e","added_by":"auto","created_at":"2025-07-08 12:02:03","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":112421,"visible":true,"origin":"","legend":"\u003cp\u003eOperating interface for dynamic nomograms predicting ALK status and Ki67 expression in patients with non-small cell lung cancer. After uploading a participant’s CT images and tumor annotations and entering age, gender, maximum diameter of the tumor size, location, pathological type, the clinician can get the participant’s corresponding probability of ALK status and Ki67 expression.\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-6829111/v1/dbedc33eb1e96e0bf9000be1.png"},{"id":86246845,"identity":"d810288c-fa81-4628-84d9-488400e57e97","added_by":"auto","created_at":"2025-07-08 11:54:03","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":1287797,"visible":true,"origin":"","legend":"\u003cp\u003eThe contribution heatmap of top 20 features’ contribution to the first 20 components after the PCA operation.\u003c/p\u003e","description":"","filename":"image5.png","url":"https://assets-eu.researchsquare.com/files/rs-6829111/v1/416542d39b58d36d7f90d649.png"},{"id":86248406,"identity":"6e4a5f48-024e-441d-8eec-01d7c136f159","added_by":"auto","created_at":"2025-07-08 12:10:11","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2661561,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6829111/v1/b01c6712-b121-4470-9473-49a3957bef12.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Multitask Radiomics-based Dynamic Nomogram for Joint Prediction of ALK Mutation and Ki-67 Expression in NSCLC","fulltext":[{"header":"Key Points","content":"\u003cul start=\"50\"\u003e\n \u003cli\u003e\u003cstrong\u003eT\u003c/strong\u003ehis study developed a multitask radiomics nomogram using preoperative CT images to simultaneously predict ALK mutation status and Ki-67 expression in NSCLC patients, achieving exceptional accuracy (AUC \u0026gt;0.96) and overcoming limitations of traditional single-task models.\u003c/li\u003e\n \u003cli\u003eThe model outperformed clinical predictors, demonstrating clinical utility through improved reclassification and decision curve analysis, offering a noninvasive alternative to tissue biopsy.\u003c/li\u003e\n \u003cli\u003eA web-based interactive platform enables real-time prediction updates, while PCA-revealed feature hierarchies highlight biologically plausible radiomic drivers, bridging the gap between artificial model and clinical interpretability.\u003c/li\u003e\n\u003c/ul\u003e"},{"header":"Introduction","content":"\u003cp\u003eNon-small-cell lung cancer (NSCLC) is one of the most frequent cancer types and is responsible for most cancer deaths worldwide\u003csup\u003e[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]\u003c/sup\u003e. Despite advances in therapy, many NSCLC patients experience relapses\u003csup\u003e[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]\u003c/sup\u003e, highlighting the need for personalized treatment strategies\u003csup\u003e[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]\u003c/sup\u003e. Biomarkers such as Anaplastic lymphoma kinase (ALK) mutation and Ki-67 expression play pivotal roles in prognosis and clinical decision-making in NSCLC\u003csup\u003e[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eALK rearrangement occur in 4\u0026ndash;5% of NSCLC patients. Up to 50\u0026ndash;60% of ALK-positive patients are at high risk for brain metastases\u003csup\u003e[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]\u003c/sup\u003e. Targeted therapies have shown efficacy in the treatment of oncogene-dependent NSCLC, which has the potential to improve outcomes\u003csup\u003e[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]\u003c/sup\u003e. Meanwhile, Ki-67, a marker of cellular proliferation, correlates with aggressive tumor behavior and poor prognosis\u003csup\u003e[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]\u003c/sup\u003e. Currently, these biomarkers are assessed via invasive tissue biopsies, which pose challenges such as sampling bias, procedural risks, and delayed results. Given the clinical importance of both biomarkers, a simultaneous prediction method could provide comprehensive insights for treatment planning.\u003c/p\u003e\u003cp\u003eComputed tomography (CT) is the primary imaging modality for NSCLC evaluation. Radiomics, a high-throughput feature extraction technique, has shown promise in noninvasively characterizing tumor heterogeneity. While prior studies have focused on single biomarkers\u003csup\u003e[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]\u003c/sup\u003e, a comprehensive consideration of multiple biomarkers is necessary in clinics. It has been challenging to combine ALK mutational status and Ki-67 expression prediction, which were traditionally treated as separate problems in medical image analysis. A multitask approach could enhance predictive accuracy by leveraging shared radiomics features between ALK and Ki-67.\u003c/p\u003e\u003cp\u003eIn this study, we proposed a multitask radiomics model to simultaneously predict ALK mutation and Ki-67 expression. By integrating clinical and radiomics data, our dynamic nomogram aims to provide a practical, noninvasive tool for preoperative biomarker assessment.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eStudy Participants\u003c/h2\u003e\u003cp\u003eEthical approval of this study was obtained from the Ethics Committees of Zhujiang Hospital and conducted in accordance with the ethical principles for medical research involving human subjects as described in the 1964 Helsinki Declaration and its later amendments. The requirement for informed consent was waived due to the retrospective nature of this study.\u003c/p\u003e\u003cp\u003eThis retrospective study included 534 NSCLC patients between January 2015 and December 2020 who met the following inclusion and exclusion criteria were enrolled in this study. Inclusion criteria were as follows: (a) pathologically confirmed NSCLC (surgical resection or biopsy); (b) preoperative CT images were available. Exclusion criteria were as follows: (a) prior chemotherapy/radiotherapy (n\u0026thinsp;=\u0026thinsp;17); (b) multiple primary carcinomas or concurrent malignancies (n\u0026thinsp;=\u0026thinsp;23); (c) poor quality of CT images (n\u0026thinsp;=\u0026thinsp;34); (d) incomplete clinical data (n\u0026thinsp;=\u0026thinsp;15). We randomly divided the patients into a training cohort and a test cohort through the 7:3 division. The statistical analysis of clinical characteristics between the two groups was shown in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eClinical characteristics of patients in training and test cohort\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eTraining cohort (n\u0026thinsp;=\u0026thinsp;373)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eTest cohort (161)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eP value\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAge (years)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e60.98 (14\u0026thinsp;~\u0026thinsp;88)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e60.11 (27\u0026thinsp;~\u0026thinsp;85)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.3754\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGender\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.2545\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e234 (62.73%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e110 (68.32%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFemale\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e139 (37.27%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e51 (31.68%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSmoking status\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.5022\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNo\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e242 (64.88%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e110 (68.32%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e131 (35.12%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e51 (31.68%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDiameter\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e36.69 (3.44\u0026thinsp;~\u0026thinsp;144.90)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e38.21 (10.00\u0026thinsp;~\u0026thinsp;139.50)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.4665\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLocation\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.6865\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLUL\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e108 (28.95%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e53 (32.92%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLLL\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e51 (13.67%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e24 (14.91%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRUL\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e116 (31.10%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e108(32.2)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRML\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e25 (6.70%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e51 (31.68%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRLL\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e73 (19.57%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e25 (15.53%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePathological type\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.6005\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSCC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e50 (13.40%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e25 (15.53%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eADC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e297 (79.62%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e122 (75.78%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eOther\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e26 (6.97%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e14 (8.70%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eALK\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.9186\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eNegative\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e337 (90.35%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e145 (90.06%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePositive\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e36 (9.65%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e16 (9.94%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eKi67 expression\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.4454\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLow\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e254 (68.10%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e115 (71.43%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHigh\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e119 (31.90%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e46 (28.57%)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"4\"\u003eValues are n (%) or mean (SD). P* were calculated by independent samples T test for Age/Diameter and Chi-square test for Gender/Smoking status/Location/Pathological type. RUL, right upper lobe; RML, right middle lobe; RLL, right lower lobe; LUL, left upper lobe; LLL, left lower lobe; GGO, ground-glass opacity; SCC, squamous cell carcinoma; ADC, adenocarcinoma.\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eClinical and Pathological Data\u003c/h3\u003e\n\u003cp\u003eBaseline clinical information was retrieved from the institutional database for medical records, including lesion density, smoking status, the maximum tumor diameter, tumor location, age, gender, and pathological type. ALK status was determined via immunohistochemistry (IHC, Ventana Benchmark XT), with positivity defined as strong cytoplasmic staining. Ki-67 expression was assessed using the CONFIRM anti-Ki-67 antibody, with high expression defined as \u0026gt;\u0026thinsp;30% nuclear staining.\u003c/p\u003e\n\u003ch3\u003eImage Acquisition and Segmentation\u003c/h3\u003e\n\u003cp\u003eCT scans were performed using Philips Brilliance iCT 256 or 64-slice scanners (120 kVp; field of view: 350mm\u0026times;350mm; matrix, 512\u0026times;512; slice thickness: 5mm; slice thickness: 5 mm). All CT images were obtained from the picture archiving and communication system (PACS). Tumor segmentation was manually performed by two radiologists (8- and 5-year experience in chest CT interpretation) using ITK-SNAP(version 3.8.0; \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://www.itksnap.org/\u003c/span\u003e\u003cspan address=\"http://www.itksnap.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e).\u003c/p\u003e\n\u003ch3\u003eRadiomics Feature Extraction and Model Development\u003c/h3\u003e\n\u003cp\u003eA total of 1437 IBSI-standardized radiomic features were extracted using the Standardized Environment for Radiomics Analysis (SERA). including: 771 first-order features (e.g. morphology, statistical, histogram and intensity-histogram features), 666 higher-order features (Supplementary table1). Besides, 6 clinical features are included (age, gender, smoke, pathological type, location, diameter).\u003c/p\u003e\u003cp\u003eAfter standardizing the features using Z-score normalization to ensure uniform scale across all dimensions, variance thresholding was applied as a basic feature selection method to remove low-variance features (threshold\u0026thinsp;=\u0026thinsp;0.01), aiming to eliminate redundant or nearly constant features. Subsequently, Principal Component Analysis (PCA) was performed to further reduce dimensionality by transforming the data into a lower-dimensional space that retains 95% of the original variance, with additional analysis conducted to examine the explained variance ratio per principal component and the most influential original features contributing to each component. In the end, the feature dimension was reduced to 63.\u003c/p\u003e\u003cp\u003eAfter that, we built the multi-task learning framework for the ALK and Ki-67 classification, where a base classifier \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\mathcal{M}}_{\\theta\\:}\\)\u003c/span\u003e\u003c/span\u003e is extended to handle multivariate target predictions through the MultiOutputClassifier formalism. Assuming there are \u003cem\u003et\u003c/em\u003e tasks (t\u0026thinsp;=\u0026thinsp;2), features \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:X\\in\\:{\\mathcal{R}}^{N*D}(\\text{D}=63)\\)\u003c/span\u003e\u003c/span\u003e, and labels \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:Y\\in\\:{\\mathcal{R}}^{N*K}(K=2)\\)\u003c/span\u003e\u003c/span\u003e, the composite model \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{\\mathcal{F}}_{\\varTheta\\:}\\)\u003c/span\u003e\u003c/span\u003e is formulated as:\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:{\\mathcal{F}}_{\\varTheta\\:}\\left(X\\right)\\:=\\:\\left[\\sigma\\:\\right({\\mathcal{M}}_{{\\theta\\:}_{1}}\\left(X\\right)),\\:\\sigma\\:({\\mathcal{M}}_{{\\theta\\:}_{2}}\\left(X\\right))\\:,...,\\sigma\\:({\\mathcal{M}}_{{\\theta\\:}_{t}}\\left(X\\right)\\left)\\:\\right]$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eWhere, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\sigma\\:\\)\u003c/span\u003e\u003c/span\u003e denotes the task decision model. Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e depicts the workflow of the proposed model.\u003c/p\u003e\n\u003ch3\u003eEvaluation of Model Performance\u003c/h3\u003e\n\u003cp\u003eThe receiver operating characteristic curves (ROCs) of the combined model and clinical model in both the training and validation groups were plotted, and the diagnostic accuracy, sensitivity, and specificity were calculated to evaluate the discriminative performance of the models. \u003cem\u003eDelong\u003c/em\u003e\u0026rsquo;s test was used to compare the ROC curves of models. The performance improvement introduced by the inclusion of multitask radiomics signature was quantified by net reclassification improvement (NRI). Furthermore, a decision curve analysis (DCA) was used to estimate clinical usefulness.\u003c/p\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003eStatistical Analysis\u003c/h2\u003e\u003cp\u003eWe performed a comparison for each clinical information between the training group and validation group for NSCLC patients. Categorical variables were analyzed by either Chi-square test. Continuous variables were compared between groups using a student\u0026rsquo;s t-test for independent samples. All analyses were performed using Python 3.8 platform. All tests for statistical significance were two-tailed and P values less than 0.05 were considered statistically significant.\u003c/p\u003e\u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\u003ch2\u003eClinical Characteristics\u003c/h2\u003e\u003cp\u003eAll patients were randomly allocated to the training cohort and validation cohort at a ratio of 7:3. Pathology confirmed 36 cases of ALK-positive (9.65%) in training cohort and 16 cases (9.94%) in validation cohort (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). 119 cases showed high Ki-67 expression (31.90%) in training cohort and 46 cases (28.57%) in validation cohort (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003eModel Performance in Discriminating ALK Mutation and Ki-67 Expression\u003c/h2\u003e\u003cp\u003eThe multitask integrative model showed better discrimination accuracy than clinical model (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, \u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). The multitask integrative model yielded the higher AUC (0.9918 [95%CI: 0.9744-1.0000]) in predicting the ALK status compared with clinical model (0.5434 [95%CI: 0.3424\u0026ndash;0.7249]). Meanwhile, The AUC of multitask integrative model for predicting the Ki67 expression was 0.9807 [95%CI: 0.9602\u0026ndash;0.9950], which was superior to the clinical model (0.7044 [95%CI: 0.6229\u0026ndash;0.7814]).\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eThe predictive performance of the multi-task model and clinical model for ALK mutation in training cohort and test cohort\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"7\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e\u003cp\u003eTraining cohort\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e\u003cp\u003eTest cohort\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMulti-task model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSingle-task model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eP value\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eMulti-task model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eClinical model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003eP value\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAUC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.9974\u003c/p\u003e\u003cp\u003e[95%CI: 0.9923, 1.0000]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.9993\u003c/p\u003e\u003cp\u003e[95%CI: 0.9973-1.0000]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.18\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.9918\u003c/p\u003e\u003cp\u003e[95%CI: 0.9744-1.0000]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.5434\u003c/p\u003e\u003cp\u003e[95%CI: 0.3424\u0026ndash;0.7249]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.05\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAccuracy\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.9866\u003c/p\u003e\u003cp\u003e[95%CI: 0.9759\u0026ndash;0.9973]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.9783\u003c/p\u003e\u003cp\u003e[95%CI: 0.9625\u0026ndash;0.9920]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.9689\u003c/p\u003e\u003cp\u003e[95%CI: 0.9379\u0026ndash;0.9938]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.6282\u003c/p\u003e\u003cp\u003e[95%CI: 0.5528\u0026ndash;0.7019]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSensitivity\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.9720\u003c/p\u003e\u003cp\u003e[95%CI: 0.9062-1.0000]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1.0000\u003c/p\u003e\u003cp\u003e[95%CI: 1.0000\u0026ndash;1.0000]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.9386\u003c/p\u003e\u003cp\u003e[95%CI: 0.7857-1.0000]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.5691\u003c/p\u003e\u003cp\u003e[95%CI: 0.2857\u0026ndash;0.8182]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSpecificity\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.9882\u003c/p\u003e\u003cp\u003e[95%CI: 0.9763\u0026ndash;0.9971]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.9760\u003c/p\u003e\u003cp\u003e[95%CI: 0.9592\u0026ndash;0.9910]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.9722\u003c/p\u003e\u003cp\u003e[95%CI: 0.9437\u0026ndash;0.9932]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.6346\u003c/p\u003e\u003cp\u003e[95%CI: 0.5555\u0026ndash;0.7114]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eThe predictive performance of the multi-task model and clinical model for Ki-67 expression in training cohort and test cohort\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"7\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e\u003cp\u003eTraining cohort\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e\u003cp\u003eTest cohort\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMulti-task model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eClinical model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eP value\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eMulti-task model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eClinical model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003eP value\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAUC\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.9778\u003c/p\u003e\u003cp\u003e[95%CI: 0.9648\u0026ndash;0.9878]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.9899\u003c/p\u003e\u003cp\u003e[95%CI: 0.9820\u0026ndash;0.9957]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.05\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.9807\u003c/p\u003e\u003cp\u003e[95%CI: 0.9602\u0026ndash;0.9950]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.7044\u003c/p\u003e\u003cp\u003e[95%CI: 0.6229\u0026ndash;0.7814]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e\u0026lt;\u0026thinsp;0.05\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAccuracy\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.9033\u003c/p\u003e\u003cp\u003e[95%CI: 0.8740\u0026ndash;0.9303]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.9409\u003c/p\u003e\u003cp\u003e[95%CI: 0.9142\u0026ndash;0.9651]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.9187\u003c/p\u003e\u003cp\u003e[95%CI: 0.8758\u0026ndash;0.9565]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.6588\u003c/p\u003e\u003cp\u003e[95%CI: 0.5839\u0026ndash;0.7267]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSensitivity\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.8907\u003c/p\u003e\u003cp\u003e[95%CI: 0.8349\u0026ndash;0.9440]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.9490\u003c/p\u003e\u003cp\u003e[95%CI: 0.9065\u0026ndash;0.9835]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.9119\u003c/p\u003e\u003cp\u003e[95%CI: 0.8205\u0026ndash;0.9796]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.7401\u003c/p\u003e\u003cp\u003e[95%CI: 0.6052\u0026ndash;0.8628]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSpecificity\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.9091\u003c/p\u003e\u003cp\u003e[95%CI: 0.8740\u0026ndash;0.9425]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.9371\u003c/p\u003e\u003cp\u003e[95%CI: 0.9057\u0026ndash;0.9656]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.9213\u003c/p\u003e\u003cp\u003e[95%CI: 0.8700-0.9658]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.6263\u003c/p\u003e\u003cp\u003e[95%CI: 0.5366\u0026ndash;0.7155]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003eFeature Preprocessing Strategy Analysis\u003c/h2\u003e\u003cp\u003eThe ROC curves of the multitask integrative model to predict the ALK status and Ki67 expression using original features and processed features are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. After computing the prior components using PCA analysis, model performance obtained a significant improvement in both training cohort and test cohort. It may be due to the ability in avoiding the negative effects of some noisy features. Importantly, the gains were consistently large in both data groups, indicating that the performance gains were mostly due to the powerful feature representations which lead to better convergence.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003eNet Reclassification Improvement Analysis of Our method and Clinical Model\u003c/h2\u003e\u003cp\u003eThe Net Reclassification Improvement (NRI) results in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e demonstrate that our method achieves significant performance gains over clinical model particularly on the test cohort. For the ALK biomarker, the test set NRI of 0.727 indicates a substantial improvement in classification accuracy, with 0.438 (NRI+) reflecting enhanced classification ability of positive cases and 0.289 (NRI-) showing improved downward reclassification of negative cases. It means that the performance gains majorly come from the correction of positive cases. Similarly, the Ki-67 biomarker exhibits a test NRI of 0.387, driven by complementary gains in both NRI+ (0.109) and NRI- (0.278). Rather, the performance gains of Ki-67 expression is due to the correction of negative cases.\u003c/p\u003e\u003cp\u003e\u003cimg src=\"data:image/png;base64,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\" width=\"584\" height=\"157\"\u003e\u003c/p\u003e\u003cp\u003eIn contrast, the training data shows mixed results, with ALK (NRI = -0.040) and Ki67 (NRI = -0.139) displaying minor degradation in reclassification performance. This discrepancy between training and test metrics may arise from regularization effects or overfitting mitigation in our framework, which prioritizes generalization over training data alignment. These findings underscore the value of our approach in enhancing ALK and Ki-67 prediction reliability at the same time, particularly for such imbalanced classification tasks.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003eDecision Curve Analysis of Our Prediction Model\u003c/h2\u003e\u003cp\u003eThe Decision Curve Analysis (DCA) in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e demonstrates that our model provides significant clinical utility for both ALK and Ki-67 predictions. For ALK, the model outperforms \"treat all\" and \"treat none\" strategies across most threshold probabilities (0.2\u0026ndash;0.8), with peak net benefit at intermediate thresholds (0.3\u0026ndash;0.6), indicating optimal decision-making for balancing over-treatment avoidance and sensitivity in clinical practice. For Ki-67, the model shows superior net benefit at low thresholds (\u0026lt;\u0026thinsp;0.4), highlighting its value in early risk identification for high-risk patients, while maintaining utility even at higher thresholds. These results suggest that integrating our model into clinical workflows could improve personalized risk stratification, reduce unnecessary interventions.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\u003ch2\u003eOperating System of Our Multi-task Prediction Model\u003c/h2\u003e\u003cp\u003eWe built an operating interface for dynamic nomograms predicting ALK status and Ki67 expression in patients as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e. It comprises the input groups on the left and the output groups on the right. Operators can upload a participant\u0026rsquo;s CT Images and the tumor delineations at first and the system will automatically compute the radiomics fast. Then, operators could enter the clinical features, i.e., age, gender, pathological type, tumor size and location in this interface. The corresponding probability and 95% confidence intervals of ALK status and Ki67 expression will be shown in the right.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\u003ch2\u003ePCA-Revealed Feature Contribution Analysis\u003c/h2\u003e\u003cp\u003eThe heatmap analysis (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e) of PCA-derived feature contributions reveals a hierarchical structure in feature importance, quantifying the top 20 features' loadings on the first 20 principal components (PCs). A subset of radiomic features, including original_glm_ZoneEntropy and logarithm_ngth_Correlation, dominates variance explanation in early PCs (PC1\u0026ndash;PC5), which primarily capture global textural descriptors. Distinct contribution patterns are observed: features such as wavelet-LH1_LLL_glm_ClusterProminence exhibit consistent high positive or negative loadings (\u0026plusmn;\u0026thinsp;0.05\u0026ndash;\u0026plusmn;0.10) across multiple PCs, indicating their robustness in capturing orthogonal data variations. Conversely, features like wavelet-LLL_glcm_ClusterShade show negligible contributions (\u0026thinsp;\u0026lt;\u0026thinsp;\u0026plusmn;\u0026thinsp;0.02), suggesting redundancy in the reduced-dimensional space. Later PCs (PC10\u0026ndash;PC20) reflect specialized variations driven by features such as lbp-3D-k_glrlm_RunLengthNoUniformityNormalized. This stratified feature contribution pattern demonstrates PCA's capacity to prioritize informative features for downstream modeling while suppressing noise.\u003c/p\u003e\u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe identification of biomarkers is important for NSCLC patients to determine the suitability of treatment and therapeutic outcomes, which is critical in precision medicine for NSCLC patients\u003csup\u003e[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]\u003c/sup\u003e. Due to the limitations of invasive biopsy or surgery, constructing an easy-to-use model based on easily available noninvasive clinical information has become a concern. In this study, the multitask integrative models were built as noninvasive tools to determine the ALK mutation and Ki-67 expression status in patients with NSCLC. The present results suggested that it is feasible to use multitask machine learning based model to predict multiple clinical indicators simultaneously.\u003c/p\u003e\u003cp\u003eOur results showed that ALK mutation was common in younger patients and male patients with adenocarcinoma are more prone to high Ki-67 expression. Our findings align with previous studies showing the predictive value of radiomics for ALK mutation, while extending the approach to include Ki-67 expression\u003csup\u003e[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e. Although the above factors played an important role in the identification of ALK mutation and Ki-67 expression, the clinical model constructed with these factors had a relatively low AUC value. Compared with the clinical model, multitask integrative models improve the prediction of ALK mutation and Ki-67 expression simultaneously. Our study found that the multitask integrative model can predict ALK mutation in NSCLC, with an AUC of 0.9918 in validation cohort. Tan et al\u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e built a model to distinguish EGFR mutant status and ALK rearrangement concurrently with an AUC of 0.70, possibly because of the mutually exclusive status between EGFR and ALK alterations. The assessment of ALK mutation and Ki-67 expression is important for prognosis and treatment management\u003csup\u003e[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]\u003c/sup\u003e. Ki-67 is a nuclear protein that is associated with cellular proliferation, conveying significant prognostic and predictive value in a variety of solid tumors\u003csup\u003e[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]\u003c/sup\u003e. ALK could stimulate cell proliferation and inhibit apoptosis. ALK mutated tumors display more aggressive behavior\u003csup\u003e[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]\u003c/sup\u003e. The inter-task correlation allows multitask machine learning exploits similar and complementary evaluation of tumor characteristics, hence improving the classification performance of model. Ganeshan et al\u003csup\u003e[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]\u003c/sup\u003e found that the texture features of NSCLC on CT were significantly correlated with the expression of tumor glucose transporter 1 (Glut-1) and tumor CD34, indicating that the CT radiomics features of the lesion reflected tumor angiogenesis and histopathological changes, such as hypoxia. However, those markers indirectly reflect tumor proliferation. In our study, machine learning aided radiomics approaches were applied to detect the Ki-67 expression, which can directly reflect tumor proliferation.\u003c/p\u003e\u003cp\u003eMultitask algorithm has been applied in previous studies. Qi et al\u003csup\u003e[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/sup\u003e provided a deep learning approach in lung cancer for automatically fast-tracking tumor lesions and classifying corresponding histological subtypes. Dong et al\u003csup\u003e[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]\u003c/sup\u003e proposed a multitasking deep learning model for predicting EGFR and KRAS mutation simultaneously and outperform single predictions. In our study, we selected radiomics features correlated with ALK mutation and Ki-67 expression based on multitask machine learning. The shared feature selection process may capture common tumor characteristics relevant to both biomarkers to improve robustness.\u003c/p\u003e\u003cp\u003eOur method is a non-invasive auxiliary detection method suitable for avoiding invasive damage when surgery and biopsy are not convenient. Also, CT images are easily available throughout the treatment period to monitor ALK mutation and Ki-67 expression status. The acquisition of CT images is relatively inexpensive in terms of cost and time. Moreover, our method does not require physicians\u0026rsquo; domain knowledge, facilitating the economical and convenient prediction of ALK mutation and Ki-67 expression.\u003c/p\u003e\u003cp\u003eHowever, the present study has certain limitations. First, this retrospective study was conducted in one center. Ideally, a prospective multicenter study would enhance the conclusion of this study. Further research is necessary to test the generalizability of our models in other populations. Second, statistical modeling based on radiomics requires large samples to obtain optimized classifiers for prediction, but our sample size was not large enough. Third, more clinical characteristics should be included to strengthen the performance and outcomes of this study. Therefore, additional prospective data collection should be involved in future study.\u003c/p\u003e\u003cp\u003eIn conclusion, the proposed nomogram could serve as a preoperative decision-support tool, enabling personalized treatment planning without invasive procedures.\u0026rdquo;\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003e\u003cstrong\u003eNSCLC\u003c/strong\u003e\u0026nbsp; \u0026nbsp; Non-small cell lung cancer\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eALK\u0026nbsp;\u003c/strong\u003eAnaplastic lymphoma kinase\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTKIs\u003c/strong\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; Tyrosine kinase inhibitors\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eIHC\u003c/strong\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;Immunohistochemistry\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eROI\u003c/strong\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;Region of interest\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eICC\u003c/strong\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;Interclass correlation coefficients\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSERA\u0026nbsp;\u003c/strong\u003eStandardized Environment for Radiomics Analysis\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eIBSI\u0026nbsp;\u003c/strong\u003eImage Biomarker Standardization Initiative\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eROCs\u003c/strong\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp;Receiver operating characteristic curves\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eNRI\u0026nbsp;\u003c/strong\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; Net reclassification improvement\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDCA\u003c/strong\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; Decision curve analysis\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eEthical approval of this study was obtained from the Ethics Committees of Zhujiang Hospital.\u0026nbsp;The requirement for informed consent was waived due to the retrospective nature of this study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe requirement for informed consent was waived due to the retrospective nature of this study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data that support the findings of this study are available on request from the corresponding author upon reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNone.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was financially supported by Science and Technology Program of Guangzhou (2024A04J5118), and President Foundation of Zhujiang Hospital, Southern Medical University (yzjj2022qn19, yzjj2022qn33).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026rsquo; contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTing Lin: Writing \u0026ndash; original draft, Visualization, Methodology, Investigation, Funding acquisition. Shengwang Peng: Writing \u0026ndash; original draft, Visualization, Methodology, Investigation. Chuanyan Li: Methodology, Investigation. Yue Yu: Methodology, Investigation. Muyao Zhou: Investigation. Xianyue Quan: Methodology, Investigation. Xin Chen: Writing \u0026ndash; review \u0026amp; editing, Supervision. Jianbing Zhu: Writing \u0026ndash; review \u0026amp; editing, Funding acquisition, Conceptualization. Dong Zeng: Writing \u0026ndash; review \u0026amp; editing, Supervision, Resources, Conceptualization.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors acknowledge all contributors who assisted in this study.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eHendriks LEL, Remon J, Faivre-Finn C, Garassino MC, Heymach JV, Kerr KM, et al. Non-small-cell lung cancer. Nat reviews Disease primers. 2024;10(1):71.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChang JY, Lin SH, Dong W, Liao Z, Gandhi SJ, Gay CM, et al. Stereotactic ablative radiotherapy with or without immunotherapy for early-stage or isolated lung parenchymal recurrent node-negative non-small-cell lung cancer: an open-label, randomised, phase 2 trial. Lancet. 2023;402(10405):871\u0026ndash;81.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang M, Herbst RS, Boshoff C. Toward personalized treatment approaches for non-small-cell lung cancer. Nat Med. 2021;27(8):1345\u0026ndash;56.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePike LRG, Miao E, Boe LA, Patil T, Imber BS, Myall NJ, et al. Tyrosine Kinase Inhibitors With and Without Up-Front Stereotactic Radiosurgery for Brain Metastases From EGFR and ALK Oncogene-Driven Non-Small Cell Lung Cancer (TURBO-NSCLC). J Clin Oncol. 2024;42(30):3606\u0026ndash;17.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYang Y, Shao X, Li Z, Zhang L, Yang B, Jin B, et al. Prognostic heterogeneity of Ki67 in non-small cell lung cancer: A comprehensive reappraisal on immunohistochemistry and transcriptional data. J Cell Mol Med. 2024;28(14):e18521.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNagpal S, Milano MT, Chiang VL, Soltys SG, Brackett A, Halasz LM, et al. Executive summary of the American Radium Society appropriate use criteria for brain metastases in epidermal growth factor receptor mutated-mutated and ALK-fusion non-small cell lung cancer. Neurooncology. 2024;26(7):1195\u0026ndash;212.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMeyer ML, Fitzgerald BG, Paz-Ares L, Cappuzzo F, J\u0026auml;nne PA, Peters S, et al. New promises and challenges in the treatment of advanced non-small-cell lung cancer. Lancet. 2024;404(10454):803\u0026ndash;22.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAtari M, Imai K, Nanjo H, Wakamatsu Y, Takashima S, Kurihara N, et al. Rapid intraoperative Ki-67 immunohistochemistry for lung cancer using non-contact alternating current electric field mixing. Lung cancer (Amsterdam Netherlands). 2022;173:75\u0026ndash;82.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePrelaj A, Miskovic V, Zanitti M, Trovo F, Genova C, Viscardi G, et al. Artificial intelligence for predictive biomarker discovery in immuno-oncology: a systematic review. Annals oncology: official J Eur Soc Med Oncol. 2024;35(1):29\u0026ndash;65.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChen M, Lu H, Copley SJ, Han Y, Logan A, Viola P, et al. A Novel Radiogenomics Biomarker for Predicting Treatment Response and Pneumotoxicity From Programmed Cell Death Protein or Ligand-1 Inhibition Immunotherapy in NSCLC. J Thorac Oncol. 2023;18(6):718\u0026ndash;30.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRiely GJ, Wood DE, Ettinger DS, Aisner DL, Akerley W, Bauman JR, et al. Non-Small Cell Lung Cancer, Version 4.2024, NCCN Clinical Practice Guidelines in Oncology. J Natl Compr Cancer Network: JNCCN. 2024;22(4):249\u0026ndash;74.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTan AC, Tan DSW. Targeted Therapies for Lung Cancer Patients With Oncogenic Driver Molecular Alterations. J Clin Oncol. 2022;40(6):611\u0026ndash;25.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGu Q, Feng Z, Liang Q, Li M, Deng J, Ma M, et al. Machine learning-based radiomics strategy for prediction of cell proliferation in non-small cell lung cancer. Eur J Radiol. 2019;118:32\u0026ndash;7.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTan X, Li Y, Wang S, Xia H, Meng R, Xu J, et al. Predicting EGFR mutation, ALK rearrangement, and uncommon EGFR mutation in NSCLC patients by driverless artificial intelligence: a cohort study. Respir Res. 2022;23(1):132.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSchneider JL, Lin JJ, Shaw AT. ALK-positive lung cancer: a moving target. Nat cancer. 2023;4(3):330\u0026ndash;43.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang J, Dong L, Zheng Z, Zhu Z, Xie B, Xie Y, et al. Effects of different KRAS mutants and Ki67 expression on diagnosis and prognosis in lung adenocarcinoma. Sci Rep. 2024;14(1):4085.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eUxa S, Castillo-Binder P, Kohler R, Stangner K, M\u0026uuml;ller GA, Engeland K. Ki-67 gene expression. Cell Death Differ. 2021;28(12):3357\u0026ndash;70.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCooper AJ, Sequist LV, Lin JJ. Third-generation EGFR and ALK inhibitors: mechanisms of resistance and management. Nat reviews Clin Oncol. 2022;19(8):499\u0026ndash;514.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGaneshan B, Goh V, Mandeville HC, Ng QS, Hoskin PJ, Miles KA. Non-small cell lung cancer: histopathologic correlates for texture parameters at CT. Radiology. 2013;266(1):326\u0026ndash;36.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eQi J, Deng Z, Sun G, Qian S, Liu L, Xu B. One-step algorithm for fast-track localization and multi-category classification of histological subtypes in lung cancer. Eur J Radiol. 2022;154:110443.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDong Y, Hou L, Yang W, Han J, Wang J, Qiang Y, et al. Multi-channel multi-task deep learning for predicting EGFR and KRAS mutations of non-small cell lung cancer on CT images. Quant imaging Med Surg. 2021;11(6):2354\u0026ndash;75.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Supplementary Table","content":"\u003cp\u003eSupplementary Table 1 is not available with this version.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-cancer","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bcan","sideBox":"Learn more about [BMC Cancer](http://bmccancer.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bcan/default.aspx","title":"BMC Cancer","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Radiomics, anaplastic lymphoma kinase, Ki-67, Non–small-cell lung cancer","lastPublishedDoi":"10.21203/rs.3.rs-6829111/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6829111/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eObjectives\u003c/h2\u003e\u003cp\u003eAnaplastic lymphoma kinase (ALK) mutation and Ki-67 expression are clinically significant biomarkers for non-small-cell lung cancer (NSCLC) assessment and treatment. Most prior studies have focused on single biomarkers while a comprehensive consideration of multiple biomarkers is necessary in clinics. The study aimed to develop a multitask radiomics-based dynamic nomogram to simultaneously predict ALK mutational status and Ki-67 expression.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e\u003cp\u003eA total of 534 patients with pathologically confirmed NSCLC (via surgical resection or needle biopsy) were retrospectively enrolled. Multitask radiomics models were developed using preoperative CT imaging features to jointly predict ALK mutation status and Ki-67 expression levels. Model performance was evaluated through area under the receiver operating characteristic curve (AUC), net reclassification improvement (NRI), and decision curve analysis (DCA). A web-based interactive platform was implemented to visualize the dynamic nomogram. Principal component analysis (PCA) was applied to identify the most influential radiomic features driving model predictions.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e\u003cp\u003eThe multitask radiomics model significantly outperformed the clinical model in validation cohorts. For ALK mutation prediction, the AUC was 0.9974 (95%CI: 0.9923, 1.0000) in the training cohort and 0.9918 (95%CI: 0.9744-1.0000) in the validation cohort. For Ki-67 expression, the AUC was 0.9778 (95%CI: 0.9648\u0026ndash;0.9878) for training and 0.9807 (95%CI: 0.9602\u0026ndash;0.9950) for testing, respectively. NRI analysis confirmed significant reclassification improvements over traditional clinical model (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05), and DCA revealed a higher net clinical benefit across clinically relevant threshold probabilities. PCA leaded to significant improvements in both training cohort and test cohort. The feature contribution analysis identified key radiomic features, such as original_glszm_zoneentropy (capturing zone size heterogeneity in original images), as dominant contributors to model predictions.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e\u003cp\u003eThe multitask radiomics-based dynamic nomogram demonstrated robust predictive performance for ALK mutation and Ki-67 expression, offering a noninvasive preoperative tool to guide personalized treatment in NSCLC.\u003c/p\u003e","manuscriptTitle":"Multitask Radiomics-based Dynamic Nomogram for Joint Prediction of ALK Mutation and Ki-67 Expression in NSCLC","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-08 11:45:51","doi":"10.21203/rs.3.rs-6829111/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewersInvited","content":"","date":"2025-07-03T11:17:49+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-06-09T16:51:42+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-06-09T01:15:33+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-06-09T01:13:18+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Cancer","date":"2025-06-05T12:20:27+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-cancer","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bcan","sideBox":"Learn more about [BMC Cancer](http://bmccancer.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bcan/default.aspx","title":"BMC Cancer","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"12fc1a6e-c4e3-4e2e-bb86-41c566071b73","owner":[],"postedDate":"July 8th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-07-08T11:45:52+00:00","versionOfRecord":[],"versionCreatedAt":"2025-07-08 11:45:51","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6829111","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6829111","identity":"rs-6829111","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}