Fig 1.
Anatomical location of the antler samples (modern corpus).
Fig 2.
Red deer antler artefacts from the Magdalenian layers (E, II, I, F1) of Isturitz cave sampled for this study.
Projectile points: 1,914. Half-round rod from tine element: 193. Unidentified rod fragments: 150, 435 and 501. Wedges from rod: 124 and 497. Flat blank: 729. Manufacturing wastes from tine (141) and beam (198) elements. Tine tip element: 132.
Fig 3.
Reindeer antler artefacts from the Magdalenian layers (E, II, I, F1) of Isturitz cave sampled for this study.
Projectile points: 97. Unidentified rod fragments: 170, 114, 333, 136, and 199. Flat blanks: 128 and 4. Manufacturing wastes from tine (5), tine tip (206), and palmation (357) elements.
Fig 4.
Image processing steps including the selection of the volume of interest (VOI) and the segmentation principle.
The 3D acquisitions were performed with an isotropic voxel size of 20x20x20 μm3 (90 kV, 160 microA, 180 s) using a high-resolution X-ray micro-CT device. The same elliptical VOI (132.9 mm3), still positioned at the same location (at the junction between the central trabecular tissue and the external compact tissue), was applied to all the 39 modern samples and 50 archaeological ones (both corpuses are presented in S1 and S3 Tables). The same global binary threshold was applied to each data stack in order to isolate the mineralised and non-mineralised structures for subsequent quantitative biometric analysis including BV/TV, Tb.N, Tb.Th, Tb.Sp, SMI and Tb.Pf.
Table 1.
Mean and standard deviation of the biometric measurements and inter-group comparisons (Wilcoxon test) for the modern corpus.
It included 39 antler samples of which 17 were red deer samples and 22 were reindeer samples (S1 Table).
Table 2.
Mean and standard deviation of the biometric measurements and inter-group comparisons (Wilcoxon test) for the archaeological corpus.
It included 50 antler artefacts of which 23 came from red deer and 27 from reindeer (S3 Table).
Fig 5.
Principle Component Analysis (PCA) showing the distribution of the modern corpus and archaeological artefacts according to Tb.N, Tb.Sp, BV/TV, Tb.Pf, Tb.Th, and SMI.
PCA (descriptive statistics) and the correlation circle established from the datasets described in S1 and S3 Tables. The separation on the factorial plane between both species is clearer for the modern corpus than for the archaeological one.
Table 3.
Confusion matrix from the SVM classification (predictive statistics) obtained by leave-one-out cross-validation on the modern corpus.
Decision rule allowing both species to be distinguished with an accuracy of 82%.
Table 4.
Confusion matrix from the SVM classification (predictive statistics) obtained by cross-validation in which the training dataset was the modern corpus and the validation dataset was the archaeological corpus.
The decision rule for modern material could not be transferred to the archaeological material (accuracy: 42%).
Table 5.
Confusion matrix from the SVM classification (predictive statistics) obtained by leave-one-out cross-validation on the archaeological corpus.
Decision rule allowing both species to be distinguished with an accuracy of 96%.