For ovarian cancer patients, paclitaxel remains to be primary chemotherapy drug. Once drug resistance is developed, it will lead to tumor progression and metastasis during chemotherapy. Many studies have shown that the development of drug resistance in cancer cells can cause morphological changes. Digital holographic microscopy is an interferometric imaging technique that can obtain 3D quantitative morphological information of label-free cells. Combining with microfluidics enables high-throughput holographic image acquisition of suspended cells. In this work, four kinds of epithelial ovarian cancer cells with different drug sensitivity, SKOV3 cells, SKOV3_Ta_2μM cells, SKOV3_Ta_8μM cells, and SKOV3_Ta_20μM cells were studied. Several machine learning algorithms were used to perform multi-classification on the extracted morphological features of four types of cells. Then, we employ the SHapley Additive exPlanations (SHAP) method to interpret the classification model. The SHAP value of each feature is calculated and sorted to obtain the important morphological features.
|