Skip to main content
Fig. 3 | Acta Neuropathologica Communications

Fig. 3

From: Novel rapid intraoperative qualitative tumor detection by a residual convolutional neural network using label-free stimulated Raman scattering microscopy

Fig. 3

Step by step segmentation by analyzing patch-pixels. A Examples of SRH patches with metastatic tumor area (red box), a non-tumor area with reactive astrocytes (green box) and low-quality area without cells (blue box), SRH from a cervical squamous cell carcinoma brain metastasis. B Pixel-level probability heatmaps of each output class after patch-passing through all three residual CNNs in comparison. Small differences in classification at the patch size level did not affect the overall prediction output for the whole SRH image. See the left side of the asterisk (*) in the second CNN model. C The overall probability heatmaps after summing all patch predictions and mapped as a semi-transparent overlay to assist surgeon and neuropathologist for SRH image interpretation in addition to residual CNNs predictions. Scale bars = 100 μm

Back to article page

Acta Neuropathologica Communications

ISSN: 2051-5960

Contact us