Fig. 1: Workflow of the proposed MPMRecNet framework for colorectal cancer postoperative recurrence prediction.

Tumor FFPE sections are imaged with multiphoton microscopy to obtain paired TPEF/SHG images, which are preprocessed and used to train and then apply MPMRecNet for patient-level recurrence prediction; performance is evaluated, and the prediction is integrated with clinical variables to build a nomogram for clinical use.