Abstract
PURPOSE: Hirschsprung's disease (HSCR) is an intestinal disorder characterized by the absence of nerve cells in parts of the intestinal tract. The definitive diagnosis is confirmed by a full-thickness rectal biopsy to verify the absence of ganglion cells. However, incomplete removal often causes post-operative complications. To establish an optical biopsy technique for targeting mucosa with aganglionosis of HSCR and to confirm its capability by another optical imaging modality and histopathology. METHODS: Raman spectroscopy (RS) is an emerging technique in tissue diagnosis without staining that makes it possible to support conventional diagnostics and therapeutics for achieving more precise outcomes in HSCR. We demonstrate the proof-of-concept for label-free detection of the aganglionic segment in HSCR based on an RS technique in combination with fine-tuned machine learning algorithms. RESULTS: RS distinguished the characteristics of aganglionic segments in the mucosal surface of the lesion. The altered morphology was confirmed by multiphoton microscopy. In addition, discrimination models were built and evaluated by convolutional neural networks and the decision tree combined with gradient boosting framework. CONCLUSION: The proposed method and model show a high accuracy above 90% and a pseudo-blind examination involving three HSCR patients implies the feasibility for clinical application. (195 words).