A method of predicting cellular drug inhibition due to heat stress is presented. Black phosphorus nanosheets are used as photothermal agents to induce stress granule formation in tumor cells. The addition of different drugs induces different thermal stress responses. The features of single-photon images are automatically extracted and analyzed by a convolutional neural network algorithm for classification and recognition, with a maximum accuracy rate of 94.52%. Unlike traditional visual discrimination, this method realizes intelligent recognition without human intervention, providing a new model for computer-aided diagnosis with greater objectivity.