This study sought to develop an objective grading system for myelofibrosis through characterization of collagen architecture using label-free multiphoton microscopy, thereby enabling quantitative, standardized, and automated assessment. A multiphoton imaging dataset encompassing diverse myelofibrosis grades was constructed. Five collagen features were extracted utilizing the "CT-FIRE" toolbox, and a support vector machine (SVM) classifier integrating these features was developed for four-class categorization. The model demonstrated high accuracy across all myelofibrosis grades, with area under the curve (AUC) values of 0.994, 0.956, 0.940, and 1.000, and a macro-average value of 0.973, thus achieving automated grading of myelofibrosis. This study quantified collagen content and fiber morphology in human bone marrow tissues, which facilitates objective myelofibrosis grading. Moreover, it lays a foundation for future automatic and rapid grading of myelofibrosis, which can subsequently be applied to the assessment of fibrosis in other organs.
Yu et al. (Wed,) studied this question.