AI and Cancer Research ── Kidney ── 2026-11-05
Kidney cancer: treatment, trials and AI
- Abstract
To evaluate whether general-purpose vision-language models can generate numerical features from preoperative noncontrast CT for predicting postoperative pathological AJCC stage groups in renal cell carcinoma. This retrospective study included 260 public development cases and 183 external cases from Shanxi Bethune Hospital. We evaluated 2,160 combinations of five VLMs, image inputs, radiomic retrieval, feature definitions, vector lengths and classifiers. Configuration selection used mean three-fold F1, followed by OOF threshold selection and refitting on the complete development cohort. Comparators included 3D ResNet-18, two radiomic classifiers, a retrieval classifier, MedGemma and physicians. Patient-level stratified bootstrap quantified uncertainty. Physician assistance was assessed in a sequential exploratory design. All three repeated calls produced exactly identical output text and numerical values under fixed conditions. The selected GPT-5.6 Sol workflow used bilateral crops, 21 slices, labeled retrieval, 128 semantic features and TabTransformer. Mean validation-fold F1 was 0.808 ± 0.052. External F1 was 0.698 with a 95% confidence interval of 0.563–0.814, and AUROC was 0.942 with a 95% confidence interval of 0.901–0.975. External F1 was 0.708 for ResNet and 0.688 for physician consensus. The VLM-minus-ResNet F1 difference was − 0.009 with a 95% confidence interval of − 0.133 to 0.106. Assisted consensus F1 was 0.750; its change of 0.063 had a 95% confidence interval of − 0.048 to 0.175. Prompt-generated features from preoperative noncontrast CT supported external prediction of postoperative pathological AJCC stage groups in RCC. Performance reflected the full input and modeling workflow. Assisted second reading had a higher observed F1 in the exploratory sequential assessment.
Journal IF-equivalent: 3.8 (OpenAlex 2-year mean citedness, value as of 2026-11-03, retrieved 2026-11-05; not the official Clarivate IF)Reference: Guo F, Sun S, Wang Y, Li P, Zhang Y, Peng L, et al. General-purpose vision-language models as computed tomography feature encoders for identifying advanced renal cell carcinoma. BMC Med Imaging. 2026 Nov 3 [Epub ahead of print]. doi:10.1186/s12880-026-03013-8.Checked: Abstract only - Abstract
Accurate WHO/ISUP grading of clear cell renal cell carcinoma (ccRCC) is clinically important but remains time-consuming and subjective in routine practice. Quantitative pathological assessment on whole-slide images (WSIs) is difficult to perform manually. We developed an artificial intelligence (AI) system for automated WHO/ISUP grading and quantitative pathological assessment of ccRCC, aiming to establish a quantitative grading system to enhance prognostic stratification. In this retrospective multicenter study, we analyzed H&E WSIs from 1192 patients from Sun Yat-Sen Memorial Hospital (SYSMH, n = 467), the Third affiliated hospital of Sun Yat-Sen University (SYSTH, n = 241), and The Cancer Genome Atlas (TCGA, n = 484). The AI pipeline performed tumor segmentation, segmentation of high-grade dedifferentiation (sarcomatoid and rhabdoid), and quantification of G1-G3 nuclei, generating slide-level WHO/ISUP grades and quantitative pathological features, including G1-G3 proportions and G4 tumor area ratio. Based on these quantitative features, we merged patient groups with similar recurrence-free survival (RFS) to develop a quantitative grading system. The quantitative prognostic grading system was developed in the pooled SYSMH–SYSTH cohort and independently validated in the TCGA cohort. In the validation sets, the AI-assisted WHO/ISUP grading model showed strong discriminatory performance for the 4-class grading task, with micro-AUCs (95% CIs) of 0.912 (0.892–0.932) in the SYSTH cohort and 0.846 (0.825–0.866) in the TCGA cohort. Quantitative pathological analysis showed that tumors with a high G2 proportion (> 46%) had RFS comparable to that of G3 tumors, whereas G4 tumors could be further divided into two prognostic groups according to the G4 tumor area ratio and high-grade dedifferentiation. Therefore, the quantitative grading system consists of four classifications: G1 + low G2 proportion group as Class 1; High G2 proportion group + G3 as Class 2; Low G4 risk group as Class 3; High G4 risk group as Class 4. The quantitative grading system showed a modest increase in discrimination over the WHO/ISUP grading system. Our AI system facilitates automated WHO/ISUP grading and quantitative pathological assessment of ccRCC on WSIs. The quantitative grading system may complement conventional WHO/ISUP grading for prognostic assessment in ccRCC.
Journal IF-equivalent: 3.9 (OpenAlex 2-year mean citedness, value as of 2026-10-09, retrieved 2026-10-10; not the official Clarivate IF)Reference: Zhang Q, Zhang L, Liao C, Yu Y, Chen Y, Lin Z, et al. Artificial intelligence-driven quantitative histologic grading and prognostic stratification for clear cell renal cell carcinoma. BMC Cancer. 2026 Nov 2 [Epub ahead of print]. doi:10.1186/s12885-026-17286-3.Checked: Abstract only