A mathematical framework for pre-mass detection of renal tissue anomalies from computed tomography scans: exploration of a biomechanical signature based on enhancement kinetics Authors: Amina El Ghazi, Abdessadek Aaroud, Abdelmajid Bessate Published: 2026 ABSTRACT AbstractBackgroundEarly detection of renal cancer remains challenging, as current imaging techniques predominantly identify tumors only after a visible mass has formed. Lesions smaller than 1 cm are particularly difficult to detect reliably using routine computed tomography (CT). This study presents a novel mathematical framework for detecting early renal tissue anomalies before visible mass formation, grounded in the hypothesis that tumorigenesis induces subtle biomechanical alterations in tissue enhancement patterns that are observable on CT imaging.MethodsWe introduced the Local Biomechanical Index (LBI), a voxel-wise metric quantifying local deviations in Hounsfield unit values relative to an intra-patient renal reference, and a volumetric enrichment ratio (Rt/g) measuring the spatial concentration of extreme LBI values. The single-phase framework was evaluated on a random sample of 20 cases from the KiTS19 dataset using fixed parameters: a 3-mm local radius, the 99th percentile threshold, a 10 mm × 10 mm × 10 mm three-dimensional (3D) window, and an enrichment threshold of 2.0. Voxel-level recall, precision, F1-score, and Rt/g were calculated against the reference tumor masks and summarized using means, standard deviations (SDs), and 95% confidence intervals (CIs).ResultsAcross the 20 cases, the mean recall was 0.575 (SD, 0.231; 95% CI: 0.474–0.676), the mean precision was 0.449 (SD, 0.292; 95% CI: 0.321–0.577), and the mean F1-score was 0.431 (SD, 0.221; 95% CI: 0.334–0.528). The mean Rt/g was 5.223 (SD, 5.427; 95% CI: 2.842–7.604). Recall exceeded 0.50 in 11 of 20 cases (55%), while Rt/g exceeded 2.0 in 14 cases (70%), indicating that candidate anomalies were frequently concentrated within tumor-associated regions. Because KiTS19 provides tumor annotations but does not annotate all other renal abnormalities, the reported precision represents agreement with the reference tumor masks rather than definitive precision for detecting all renal anomalies.ConclusionsThese findings provide preliminary evidence that early tumor-related biomechanical alterations may be detectable through spatial analysis of enhancement deviations in CT images. The proposed framework operates without tumor segmentation, relies on interpretable statistical measures, and can be fully automated. Although further validation on large longitudinal datasets is required, this approach opens a promising research direction for segmentation-free early detection of renal cancer and for integrating biomechanical signatures into radiological screening workflows.Keywords: Early cancer screening, kidney, computed tomography (CT), biomechanical signature, segmentation-free index Introduction Clinical context and diagnostic problem Kidney cancer represents a major public health burden, with high incidence and significant mortality when diagnosed at advanced stages (1). Although computed tomography (CT) is the primary diagnostic tool for kidney cancer, it is effective only once a visible mass has already formed and does not reliably identify pre-mass tissue transformations (2). In clinical practice, the diagnostic grey zone for lesions smaller than 1 cm means that most renal masses are diagnosed only when they have reached 2 cm or larger (3). Limitations of existing approaches and methodological gap Artificial intelligence (AI)-based technologies have recently shown progress in the detection and automatic segmentation of visible tumor masses, including through evaluation on reference datasets such as KiTS19 (4). However, most of these methods rely on supervised learning using expert-curated datasets in which tumor boundaries are manually and precisely delineated as ground-truth annotations. Although these annotations are not required as inputs during inference, the models learn predominantly from clearly identified and carefully segmented lesions. Consequently, their performance may decline substantially when applied to heterogeneous, unsegmented clinical CT examinations, particularly when lesions are subtle, poorly defined, or differ from those represented in the training data (4). State of the art Cancer is a major public health issue as its incidence is rising and too frequently has a poor prognosis when diagnosed late. CT is the first-line imaging modality for diagnosis but has poor utility in screening since reliable detection of very small (typically <1 cm) lesions is fundamentally compromised by the technology’s severe limitations in its capability to detect such lesions. Because of this technology-related limitation on reliable diagnosis of kidney cancer at very small sizes, the clinical implications are immediate. Ultrasound, in the current clinical setting, is unable to detect sub-centimeter lesions, and CT technology has inherent limitations in terms of sensitivity as well as poor reliability due to artifacts and misinterpretation of noise. In this context, there are clear and reliable diagnoses for lesions <5 mm and those >10 mm using both CT and magnetic resonance imaging (MRI), but the 5–10 mm range is a gray area because MRI, while much better at detection, is rarely used for screening. In this regard, it can be said that lesions less than 5 mm and greater than 10 mm are clear for diagnosis; however, those ranging between 5 and 10 mm constitute a gray zone, as the MRI, despite being more sensitive, is not often used in the screening stage. Given this reality, it’s no shock that by the time most renal masses are diagnosed in clinical practice they’re already 2 cm or larger; and prognosis is most favorable for lesions much smaller (i.e., localized) (3). In clinical practice, radiological analysis relies primarily on two-dimensional (2D) images from CT or MRI, with each slice corresponding to an isolated anatomical section. The radiologist must therefore manually examine a large volume of data, potentially reaching 500 to 1,000 slices for a complete thoraco-abdomino-pelvic CT scan, and mentally reconstruct the three-dimensional (3D) structure of the kidney from these successive slices. This procedure generates a substantial cognitive load, induces visual fatigue, and may contribute to inter-observer variability, particularly for microlesions whose interpretation remains challenging (5). The development of more precise detection methods, easy to integrate into routine medical practice and suitable for preliminary examinations, is therefore essential to improve early diagnosis. Hence, the clinical need for sensitive and early detection tools which could be easily incorporated into a busy radiological workflow still persists. This raises the possibility of an entirely new approach: what if the first observable signature were not morphological, but biomechanical? Oncophysics has discovered that the cancer transformation of a tissue is driven by early micro-architectural changes—hypercellularity, increase in extra-cellular matrix, and solid stress—which generate local compaction and change the mechanical response of a tissue many orders-of-magnitude before we can detect an invading cancer mass (6). The hypothesis is that this compaction leaves a subtle but detectable imprint during routine CT scanning, manifesting as a regional and statistically significant deviation from expected Hounsfield unit (HU) density. These biomechanical changes, including increased cellularity and solid stress, have been shown to precede morphological evidence and can be detected by techniques such as MRI elastography, which can quantify mechanical differences between renal tumors and renal parenchyma (7). CT provides indirect information on electron density, which is correlated with mass density and tissue composition (8), by measuring X-ray attenuation in tissue. We hypothesize that early mass effects on the density of tissue associated with tumor transformation might thus leave a subtle but detectable mark in local HUs when a distinct mass entity cannot be clearly seen. Here, we present a mathematical theory of Local Biomechanical Index (LBI) for “segmentation-free” screening, which formalizes this conjecture in this paper. The biomechanical hypothesis: the birth of a paradigm? This remark invites consideration of a much broader paradigm: what if the first observable signature was not morphological, but biomechanical? Work in oncophysics has demonstrated that early alterations of the tissue microenvironment (hypercellularity, compaction and mechanical stress modifications) precedes morphologically detectable transitions during tumorigenesis (6,7). The first of these is CT, which allocates tissue electron density [essentially a composite measure that correlates with mass density and chemical composition (8)] based on X-ray attenuation data. Our hypothesis in this study is that those tissue compaction events associated with early stages of tumorigenesis would then be able to stamp into locally resolved HUs but without the requirement for a morphologically defined mass as well. Proposition: a mathematical formula on the biomechanical spatial concentration Our contribution addresses this methodological gap. We propose to quantify, at every parenchymal point, the degree of deviation from biomechanical normality. The specificity is based on characterizing the local spatial spread of this deviation, at the voxel level, by an LBI, but especially through an analysis of the spatial concentration of extreme signatures. We aim to distinguish the focal cluster of tissue instability (end-product of an emerging lesion) from random physiological background, by studying the spatial location of these deviations in the renal parenchyma. Importantly, this approach is segmentation-free, as it requires neither a trained model nor prior learning of the target lesion. Contributions and article outline This is the entire formalization of that process, and proof-of-concept validation on actual data. Our contributions are as follows: The mathematically-based definition of the LBI had to be calculated per voxel, and results in a normalized relative deviation of tissue density from an intra-patient normality reference; The mathematically-based definition of the LBI had to be calculated per voxel, and results in a normalized relative deviation of tissue density from an intra-patient normality reference; We proposed the new concept of spatial concentration of biomechanical anomalies, with a core measure [tumor-to-global enrichment ratio (Rt/g)] to detect lesion foci without segmentation; We proposed the new concept of spatial concentration of biomechanical anomalies, with a core measure [tumor-to-global enrichment ratio (Rt/g)] to detect lesion foci without segmentation; The construction of a full volumetric experimental protocol (EXP01 to EXP04) that validates the method on the public KiTS19 dataset and demonstrates its ability to reveal the biomechanical signature of small tumors; The construction of a full volumetric experimental protocol (EXP01 to EXP04) that validates the method on the public KiTS19 dataset and demonstrates its ability to reveal the biomechanical signature of small tumors; Analysis of the results presenting significant enrichment in tumor areas for extreme anomalies and demonstrating that a straightforward image reconstruction algorithm is able to estimate the size of the lesion. Analysis of the results presenting significant enrichment in tumor areas for extreme anomalies and demonstrating that a straightforward image reconstruction algorithm is able to estimate the size of the lesion. The remainder of the paper is structured as follows: “Methods” section describes the 3D methodology and mathematical background. The volumetric experimental protocol and results are described in “Results” section. These results were interpreted in the “Discussion”, along with some of the limitations and future prospects, ending with a conclusive remark. We present this article in accordance with the TRIPOD reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2026-0685/rc). Methods Dataset and preprocessing Experimental validation was performed on cases from the public KiTS19 database (4). Cases were selected for their representativeness of real clinical challenges, including notably small tumors (case 00120) and complex anatomical presentations (case 00000). For each case, the 3D CT volume and its associated reference segmentation (background: 0; healthy parenchyma: 1; tumor: 2) were used. No additional preprocessing (registration, global intensity normalization) was applied, to preserve native radiomic information and test method robustness under realistic conditions. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. Link to clinical practice: from biomechanical signal to lesion characterization Translating this biomechanical signature into a clinically relevant tool requires establishing a clear bridge between the quantitative measurement of enhancement (HU) and the radiologist’s diagnostic reasoning. The added value of our approach lies precisely in this correspondence: the voxel-wise analysis of HU, formalized by the LBI index, makes it possible to capture the same pathophysiological characteristics that underpin the clinical interpretation of renal lesions. Indeed, as highlighted by numerous studies in renal imaging, diagnosis relies on the evaluation of specific enhancement profiles in multiphasic CT (9,10). For example, the absence of significant enhancement (HU <10) is the marker of simple cystic lesions (Bosniak I) (11), while rapid and intense enhancement in the corticomedullary phase (high wash-in rate) associated with rapid wash-out constitutes the typical signature of hypervascular clear cell renal cell carcinoma (ccRCC) (10,12). Conversely, weak and progressive enhancement points to a hypovascular tumor such as papillary carcinoma (9,13). Our LBI, by quantifying the deviation from normal local enhancement, operationalizes this clinical reading grid. A voxel or region with a high LBI does not signal a mere abstract “anomaly”; it objectifies an area whose enhancement kinetics resemble, for example, those of a hypervascular tumor. Thus, the conceptual framework we propose does not simply detect a statistical deviation; it provides a probabilistic map whose interpretation can be directly correlated with established diagnostic categories (cystic, hypervascular solid, hypovascular solid, etc.), thereby addressing a fundamental clinical need: moving from simple detection to assistance in early lesion characterization. By demonstrating that the spatial concentration of extreme LBI values corresponds to foci whose enhancement behavior is pathognomonic, we lay the groundwork for a tool that could, in the long term, not only alert to the presence of a pre-mass anomaly but also suggest its nature, bridging the gap between tissue physics and medical reasoning. Mathematical framework: LBI Let I(0):Ω⊂Z3→R be a 3D non-contrast CT volume, and I(p):Ω⊂Z3→R be a 3D post-contrast CT volume (phase p, e.g., nephrographic phase), where I(p)(x) represents the intensity value in HU at voxel with discrete coordinates x=(x,y,z). We define the local enhancement map ΔI(p) for phase p as: Let Ωkidney⊂Ωbe a binary 3D mask delineating the renal parenchyma of interest, obtained from the reference segmentation (labels 1 and 2). This mask contains no information about the presence or location of a tumor. Statistical modeling of healthy reference tissue enhancement A patient-specific reference model is established by calculating global enhancement statistics over the entire renal volume Ωkidney for a given phase p: where |Ωkidney| denotes the total number of voxels in the renal mask. This model (μΔ,global(p),σΔ,global(p)) represents the central distribution and variability of healthy renal tissue enhancement for the considered patient and phase, serving as the basis for robust internal normalization. Calculation of the LBI For each voxel x∈Ωkidney, we define a local neighborhood Vr(x). Typically, this is a sphere of radius r (in mm, converted to voxels according to image resolution) centered at x and intersected with Ωkidney: The local average enhancement value is calculated over this neighborhood: The fundamental LBI for phase p is then defined as a Z-score measuring the normalized deviation of local enhancement from the global healthy tissue enhancement reference: A high positive LBI indicates a region with locally higher enhancement (suggesting hypervascularity or compaction) than the average parenchyma, suggesting possible biomechanical alteration. In all our experiments, the neighborhood radius was set to r=3.0mm, and the nephrographic phase (p=N) was used as the primary phase for analysis due to its high diagnostic yield for hypovascular tumors. Analysis of spatial concentration of extreme anomalies The key innovation lies not only in calculating the LBI, but in analyzing the spatial distribution of its most extreme values. Identification of extreme anomalies We select the subset Ωextreme of voxels with the highest LBI values in the kidney. This is achieved by thresholding based on a high percentile p (for example, p=99): where Qp denotes the p-th percentile. Thus, Ωextreme contains the top (100−p) most abnormal voxels (e.g., top 1%). 3D scanner by enrichment ratio (Rt/g) We define a local volumetric window W(x) centered at voxel x, of fixed physical size (e.g., 10 mm × 10 mm × 10 mm), converted to voxel size according to the volume spacing. The local density of extreme events in this window is: The global density of extreme events in the entire kidney is: The tumor-to-global enrichment ratio at voxel x is then defined as: This ratio quantifies how much the local concentration of extreme anomalies deviates from the average concentration expected in the entire organ. Thresholding and candidate mask generation A voxel is declared candidate if its enrichment ratio exceeds a minimal threshold Rmin and if its local window contains enough renal tissue: Experimental protocol (EXP01 to EXP04) Method validation was structured as a series of progressive experiments designed to test different aspects of the biomechanical hypothesis. EXP01 (proof of concept): application of the method on a challenging case (case 00120) containing a very small tumor, to qualitatively observe LBI distribution and enrichment of extreme anomalies in the tumor region; EXP01 (proof of concept): application of the method on a challenging case (case 00120) containing a very small tumor, to qualitatively observe LBI distribution and enrichment of extreme anomalies in the tumor region; EXP02 (voxel-wise validation): quantitative analysis on a specific slice (case 00000, slice z=311) to verify that extreme voxels (Q99) localize exclusively in the tumor and calculate a first Rt/g ratio; EXP02 (voxel-wise validation): quantitative analysis on a specific slice (case 00000, slice z=311) to verify that extreme voxels (Q99) localize exclusively in the tumor and calculate a first Rt/g ratio; EXP04 (volumetric 3D scanner): extends the analysis to the entire 3D volume. Systematically computes the map of R(x) ratios for each kidney voxel, generates the candidate mask Ωcand, and evaluates overlap with ground truth tumor. This experiment provides the main performance metrics: recall (sensitivity), precision, and the global tumor ratio Rt/g=ρtum/ρcand, where ρtum is the density of candidates in the tumor and ρcand the global density of candidates. EXP04 (volumetric 3D scanner): extends the analysis to the entire 3D volume. Systematically computes the map of R(x) ratios for each kidney voxel, generates the candidate mask Ωcand, and evaluates overlap with ground truth tumor. This experiment provides the main performance metrics: recall (sensitivity), precision, and the global tumor ratio Rt/g=ρtum/ρcand, where ρtum is the density of candidates in the tumor and ρcand the global density of candidates. The EXP04 experiment constitutes the core of quantitative validation. It demonstrates the system’s ability to identify, based solely on the criterion of biomechanical spatial concentration of enhancement anomalies, regions that coincide with tumor presence. Implementation All the algorithms were run in Python (version 3.8) using common scientific libraries (NumPy and SciPy). 3D neighborhood and windowing operations were implemented in a single convolution form. The calculations were performed on a single workstation. LBI maps and candidate masks are stored in NIFTI files for subsequent visualization, analysis and processing using external software (3D Slicer, ITK-SNAP). Results EXP01: proof of concept on a small lesion Test case choice and scientific justification: to test the sensitivity of the biomechanical index under challenging conditions, we selected case_00120 from KiTS19, which contains one of the smallest annotated tumors in the entire database. This represents a particularly unfavorable case for evaluating the method’s performance under worst-case conditions. Test case choice and scientific justification: to test the sensitivity of the biomechanical index under challenging conditions, we selected case_00120 from KiTS19, which contains one of the smallest annotated tumors in the entire database. This represents a particularly unfavorable case for evaluating the method’s performance under worst-case conditions. Calculation and raw results: execution generates a file biomech_index_Zlocal.nii.gz. Observed statistics are as follows:Healthy Parenchyma: mean =−0.092, median =0.051;Tumor: mean =0.500, median =0.638. Calculation and raw results: execution generates a file biomech_index_Zlocal.nii.gz. Observed statistics are as follows: Healthy Parenchyma: mean =−0.092, median =0.051; Healthy Parenchyma: mean =−0.092, median =0.051; Tumor: mean =0.500, median =0.638. Tumor: mean =0.500, median =0.638. Analysis of high-index voxels:95th percentile threshold: 18,071 voxels above threshold, 443 in tumor (i.e., 2.5%);99th percentile threshold: 3,614 voxels above threshold, 326 in tumor (i.e., 9%);Distances: minimum distance =0 voxel. Thus, some voxels in the top 1% are located within the tumor itself. Analysis of high-index voxels: 95th percentile threshold: 18,071 voxels above threshold, 443 in tumor (i.e., 2.5%); 95th percentile threshold: 18,071 voxels above threshold, 443 in tumor (i.e., 2.5%); 99th percentile threshold: 3,614 voxels above threshold, 326 in tumor (i.e., 9%); 99th percentile threshold: 3,614 voxels above threshold, 326 in tumor (i.e., 9%); Distances: minimum distance =0 voxel. Thus, some voxels in the top 1% are located within the tumor itself. Distances: minimum distance =0 voxel. Thus, some voxels in the top 1% are located within the tumor itself. Numerical summary: the tumor represents only 0.32% of the kidney, but concentrates 9% of the most extreme anomalies. Numerical summary: the tumor represents only 0.32% of the kidney, but concentrates 9% of the most extreme anomalies. EXP02: validation on single slice (case 00000, slice z=311) Execution of the analysis script on slice z=311 of case 00000 produces the following console output: === EXPERIMENTAL LBI VALIDATION === + Q99 threshold = 1.0296067 + Total extreme voxels = 25 + Extreme voxels in tumor = 25 + Rt/g = 4.547486033075354 === END OF EXPERIMENT === On this slice, all extreme voxels are located within the tumor. EXP04: 3D scanner by enrichment ratio (complete case 00000) Experimental parameters (EXP04/EXP04E):Case: case_00000;Voxel spacing: (sx, sy, sz) = (0.5, 0.9199219, 0.9199219) mmLBI radius: r=3.0mm;Percentile: p=99;Q99=0.8414603651;3D window: WIN_MM = (10, 10, 10) mm => (wz, wy, wx) = (11, 11, 21) voxels;Enrichment threshold: Rmin=2.0;Minimum organ coverage: Mmin=200 voxels. Experimental parameters (EXP04/EXP04E): Case: case_00000; Case: case_00000; Voxel spacing: (sx, sy, sz) = (0.5, 0.9199219, 0.9199219) mm Voxel spacing: (sx, sy, sz) = (0.5, 0.9199219, 0.9199219) mm LBI radius: r=3.0mm; LBI radius: r=3.0mm; Percentile: p=99;Q99=0.8414603651; Percentile: p=99;Q99=0.8414603651; 3D window: WIN_MM = (10, 10, 10) mm => (wz, wy, wx) = (11, 11, 21) voxels; 3D window: WIN_MM = (10, 10, 10) mm => (wz, wy, wx) = (11, 11, 21) voxels; Enrichment threshold: Rmin=2.0; Enrichment threshold: Rmin=2.0; Minimum organ coverage: Mmin=200 voxels. Minimum organ coverage: Mmin=200 voxels. Obtained results:Organ voxels (Ωorg): 934,170;Tumor voxels (Ωtum): 18,491;Candidate voxels (Ωcand): 154,544;Candidate voxels belonging to tumor: 11,122;Recall (Tumor Coverage): 0.60148180196;Global Tumor Ratio Rt/g: 3.63576880973;Spatial analysis: candidate activation is mainly localized around the slices containing the tumor (e.g., z=325,350). Obtained results: Organ voxels (Ωorg): 934,170; Organ voxels (Ωorg): 934,170; Tumor voxels (Ωtum): 18,491; Tumor voxels (Ωtum): 18,491; Candidate voxels (Ωcand): 154,544; Candidate voxels (Ωcand): 154,544; Candidate voxels belonging to tumor: 11,122; Candidate voxels belonging to tumor: 11,122; Recall (Tumor Coverage): 0.60148180196; Recall (Tumor Coverage): 0.60148180196; Global Tumor Ratio Rt/g: 3.63576880973; Global Tumor Ratio Rt/g: 3.63576880973; Spatial analysis: candidate activation is mainly localized around the slices containing the tumor (e.g., z=325,350). Spatial analysis: candidate activation is mainly localized around the slices containing the tumor (e.g., z=325,350). Traceability of experimental execution EXP02 Figure 1 shows the execution trace of the EXP02 analysis script. Execution trace of the EXP02 analysis script. The colored area shows LBI values within the kidney. The red circle marks the annotated tumor region. On this slice, all 25 extreme LBI voxels were located within the tumor. LBI, Local Biomechanical Index. Multicase validation on 20 KiTS19 cases To determine whether the observations obtained from the initial illustrative cases could be reproduced across a broader sample, the complete single-phase experimental pipeline was applied to a random sample of 20 cases from the KiTS19 dataset. The same retained parameter configuration was used for all cases: a local radius of r=3mm, the 99th percentile threshold (Q99), a 10 mm × 10 mm × 10 mm 3D window, and an enrichment threshold of Rmin=2.0. For each case, the candidate mask generated without using the tumor annotation was compared with the reference tumor mask. Voxel-level recall, precision, F1-score, and the tumor-to-global enrichment ratio Rt/g were calculated. The tumor annotation was used exclusively for performance evaluation and was not involved in candidate generation. Aggregate results are presented as the mean, standard deviation, and 95% confidence interval (CI) across the 20 cases. Table 1 presents the aggregate performance of the proposed single-phase framework across 20 randomly selected KiTS19 cases. Table 1Aggregate performance of the proposed single-phase framework across 20 randomly selected KiTS19 casesMetricMeanSD95% CIRecall0.5750.2310.474–0.676Precision0.4490.2920.321–0.577F1-score0.4310.2210.334–0.528Rt/g5.2235.4272.842–7.604CI, confidence interval; Rt/g, tumor-to-global enrichment ratio; SD, standard deviation. Aggregate performance of the proposed single-phase framework across 20 randomly selected KiTS19 cases CI, confidence interval; Rt/g, tumor-to-global enrichment ratio; SD, standard deviation. The distribution of the results across cases is summarized in Table 2. Recall exceeded 0.50 in 11 of the 20 cases (55%) and exceeded 0.30 in 15 cases (75%). The enrichment ratio Rt/g exceeded 2 in 14 cases (70%) and exceeded 3 in 12 cases (60%). Table 2Distribution of recall and spatial enrichment results across the 20 KiTS19 casesCriterionValueRecall >0.5011/20 [55]Recall >0.3015/20 [75] Rt/g>2 14/20 [70] Rt/g>3 12/20 [60]Data are presented as n/N [%]. Rt/g, tumor-to-global enrichment ratio. Distribution of recall and spatial enrichment results across the 20 KiTS19 cases Data are presented as n/N [%]. Rt/g, tumor-to-global enrichment ratio. The mean number of false-positive candidate voxels, defined as candidate voxels located outside the reference tumor mask, was 3,052.7 per case (95% CI: 1,489.8–4,615.6). This is a voxel-level measure and should not be interpreted as the number of false-positive lesions or false-positive patients. Overall, the multicase evaluation confirms that the spatial enrichment signal was not restricted to the initially selected favorable examples. The mean Rt/g substantially exceeded 1, and 70% of the cases presented an enrichment ratio greater than 2, indicating that candidate anomalies were frequently concentrated within tumor-associated regions rather than being uniformly distributed throughout the kidney. Nevertheless, the moderate mean precision and F1-score, together with the variability observed across cases, demonstrate a substantial false-positive burden in the present single-phase implementation. It should also be noted that KiTS19 provides annotations for renal tumors but not for all other renal abnormalities; therefore, candidate regions outside the tumor mask are systematically counted as false positives, even though some may correspond to real but unannotated abnormalities, and the reported precision should consequently be interpreted as tumor-mask precision rather than as the definitive precision of the method for detecting all renal anomalies. These results therefore support the quantitative feasibility of the spatial enrichment principle, but they do not constitute definitive diagnostic or clinical validation. Discussion Scientific interpretation of preliminary results The results of these exploratory experiments are consistent with the working hypothesis: a local deviation in enhancement (HU), interpreted as a potential biomechanical signature, appears to concentrate nonrandomly in the tumor regions of the analyzed cases. On the slice studied in EXP02, all extreme voxels are located in the tumor, and in 3D (EXP04), the enrichment ratio Rt/g is markedly greater than 1 (3.64). These observations suggest that the concept of spatial concentration of biomechanical anomalies could be relevant for distinguishing a pathological focus from physiological background noise. Potential advantages of the conceptual approach This approach presents several notable characteristics: Independence from tumor segmentation: the approach circumvents a major limitation of supervised methods by not requiring annotation of the target lesion. It is designed to detect potential anomalies, including subtle ones. Independence from tumor segmentation: the approach circumvents a major limitation of supervised methods by not requiring annotation of the target lesion. It is designed to detect potential anomalies, including subtle ones. Interpretability: the LBI is linked to a pathophysiological property (enhancement kinetics), offering a more direct basis for interpretation than “black box” features. The Rt/g ratio provides a simple measure of anomaly “focality”. Interpretability: the LBI is linked to a pathophysiological property (enhancement kinetics), offering a more direct basis for interpretation than “black box” features. The Rt/g ratio provides a simple measure of anomaly “focality”. Intra-subject normalization: the use of global patient statistics (μglobal,σglobal) makes the method inherently robust to inter-individual and protocol variations. Intra-subject normalization: the use of global patient statistics (μglobal,σglobal) makes the method inherently robust to inter-individual and protocol variations. Automation: the pipeline, from image to candidate mask, is fully automatable. Automation: the pipeline, from image to candidate mask, is fully automatable. Potential generalization: the principle could be extended to other organs and modalities. Potential generalization: the principle could be extended to other organs and modalities. Synthesis and clarification of the study scope We have extended the validation to a random sample of 20 cases from KiTS19 [mean recall =0.575 (0.474–0.676), mean R(t/g) =5.22 (2.84–7.60)], performed a sensitivity analysis of the parameters, clarified the fundamental distinction between LBI (a mathematical deviation index without direct diagnostic value) and ΔHU (a precise physiological selector of vascular anomalies), and strengthened the discussion on robustness against protocol variations. We wish to reaffirm that the present study constitutes a proof of concept aimed at validating the quantitative core of the proposed framework using the single-phase KiTS19 database. This deliberate choice, which corresponds to the case where p=0 (i.e., the absence of a native pre-injection phase), does not undermine the generality of the mathematical formulas presented. All equations—including ΔI(p), μΔ,local(p), μΔ,global(p), LBI, ρW, and Rt/g—are defined for an arbitrary phase p and thus remain fully valid. The results obtained on KiTS19, while encouraging, do not represent final conclusions. They pave the way for the next phase of the project, which will consist of applying the complete framework including pre- and post-contrast acquisitions (the case where p corresponds to a post-injection phase with real ΔHU). This future validation on the multiphasic UCSF RMaC database will allow evaluation of the diagnostic potential of ΔHU as a precise selector, with the goal of reducing vascular false positives and improving detection performance. This subsequent work will be the subject of a distinct article. Limitations and research perspectives This proof-of-concept study, although encouraging, has limitations that open the way for numerous future works. Current methodological limitations Empirical parameterization: the parameters (r, percentile, Rmin) are set heuristically. Systematic optimization on a training set is necessary. Empirical parameterization: the parameters (r, percentile, Rmin) are set heuristically. Systematic optimization on a training set is necessary. Restricted validation: the results are limited to a few cases from KiTS19. Extensive validation on the entire database, and then on independent multicenter cohorts, is imperative to confirm the robustness and generalizability of the approach. Restricted validation: the results are limited to a few cases from KiTS19. Extensive validation on the entire database, and then on independent multicenter cohorts, is imperative to confirm the robustness and generalizability of the approach. Lack of real pre-mass detection: the validation is performed on already visible and annotated tumors. The true test of the concept would be its application to cases where no mass is yet visible, but where a lesion will develop later. This requires longitudinal data, which were not available here. Lack of real pre-mass detection: the validation is performed on already visible and annotated tumors. The true test of the concept would be its application to cases where no mass is yet visible, but where a lesion will develop later. This requires longitudinal data, which were not available here. Development perspectives These preliminary results lay the groundwork for future work: Post-processing by clustering: the application of 3D clustering algorithms (such as DBSCAN) on candidate voxels could filter out noise and provide coherent regions of interest (ROIs). Post-processing by clustering: the application of 3D clustering algorithms (such as DBSCAN) on candidate voxels could filter out noise and provide coherent regions of interest (ROIs). Optimization and large-scale validation: a rigorous evaluation on the entire KiTS19 dataset and other public databases is needed to quantify performance and optimize hyperparameters. Optimization and large-scale validation: a rigorous evaluation on the entire KiTS19 dataset and other public databases is needed to quantify performance and optimize hyperparameters. Longitudinal studies: to test the pre-mass detection capability, the ideal would be to apply the method to CT exams prior to the appearance of a tumor in patients who developed cancer. Longitudinal studies: to test the pre-mass detection capability, the ideal would be to apply the method to CT exams prior to the appearance of a tumor in patients who developed cancer. Hybridization with AI: a hybrid approach could be explored, where ROIs generated by our method serve as input to a lightweight neural network for fine classification, combining biomechanical interpretability and the power of deep learning. Hybridization with AI: a hybrid approach could be explored, where ROIs generated by our method serve as input to a lightweight neural network for fine classification, combining biomechanical interpretability and the power of deep learning. Multi-parametric exploration: other local descriptors (texture, gradient) could be integrated to enrich the biomechanical signature. Multi-parametric exploration: other local descriptors (texture, gradient) could be integrated to enrich the biomechanical signature. Conclusion and potential implications This study lays the theoretical and computational foundations for a new conceptual approach for early detection of kidney cancer. By proposing to map the deviation from “normal” biomechanics via enhancement kinetics in multiphasic CT scans, it offers a possible avenue to address the pre-mass detection challenge. Its relative algorithmic simplicity and interpretability make it potentially adaptable for future research. Our proof-of-concept work suggests that the biomechanical signature is indeed present and measurable in the analyzed tumors. The next steps, including parameter optimization, spatial filtering, and validation on large cohorts, aim to transform this conceptual framework into a more mature research tool, capable of being evaluated for potential assistance in early diagnosis. Conclusions This study presents a conceptual exploration for the early recognition of kidney cancer by CT scan. By breaking away from the morphological segmentation paradigm, we proposed a mathematical framework based on the local biomechanical signature of tissue, via the LBI and the analysis of spatial concentration of anomalies by the Rt/g ratio. The application of this framework to selected cases from the KiTS19 challenge provided an encouraging proof-of-concept: the studied tumor regions exhibit an enrichment in biomechanical anomalies and their spatial concentration correlates with the presence of the mass. The method is independent of tumor segmentation, interpretable, and based on robust intra-patient normalization. This work is a first step. It establishes the foundations for a new research avenue for pre-mass detection. Short-term perspectives include parameter optimization, integration of spatial clustering techniques to reduce false positives, and, above all, extensive validation on large cohorts, ideally longitudinal, to assess the true early detection potential of the approach. In the longer term, after rigorous validation, this type of conceptual framework could be considered as a warning module in the radiological workflow, complementing existing tools. Supplementary The article’s supplementary files as Acknowledgments We would like to thank Dr. Adil Goumri, urological surgeon, for his valuable assistance and support, and Khadija Imlahi for her continuous encouragement and support throughout this project. Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. Footnotes Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2026-0685/rc Funding: None. Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2026-0685/coif). The authors have no conflicts of interest to declare. References Associated Data Supplementary Materials The article’s supplementary files as