Deep Learning Based Classification of Prostate Tissue Microarray samples by Label-Free Non-linear Optical Microscopic Imaging.

Jackson Rodrigues^1*, Guan-Yu Zhuo¹

¹Institute of Biophotonics, National Yang Ming Chiao Tung University, Taipei, Taiwan

* Presenter:Jackson Rodrigues, email:jackson.rodrigues777@gmail.com

Prostate cancer remains one of the most prevalent malignancies among men worldwide and accurately differentiating between tumor and normal tissue continues to stand as a diagnostic challenge, particularly when relying solely on conventional histopathology. In this study, we propose a label-free, high-resolution imaging and deep learning-based classification pipeline for prostate tissue using two-photon fluorescence (TPF) microscopy and tissue microarray (TMA) samples. TPF images were acquired from prostate cores from normal, tumor, and control. Following preprocessing, deep learning-based features extracted from the penultimate layer of the MobileNetV3 Small architecture was employed. These features were subsequently classified using Support Vector Machine (SVM). A comparison of evaluation revealed that the deep learning-based approach achieved an overall classification accuracy of 91%. This efficient, label-free workflow demonstrates the promise of combining lightweight convolutional neural network architectures with multiphoton imaging for scalable, automated prostate cancer diagnosis and large-scale tissue screening.

Keywords: Two-Photon Fluorescence (TPF), Multiphoton Microscopy, Feature Extraction, Deep Learning, Support Vector Machine (SVM)