Fat　research　of　thigh　is　very　valuable　in　the　diagnosis　of　metabolic　syndrome　and　metabolic　dysfunction.　However,　it　is　more　difficult　to　segment　the　subcutaneous　fat　and　intermuscular　fat　in　MRI　images　because　of　connected　regions.　The　result　of　expectation　maximization　algorithm　is　used　to　improve　level-set　algorithm　and　realize　automatic　MRI　image　segmentation.　Results　show　that　the　subcutaneous　fat　tissue,　intermuscular　fat　tissue,　and　other　tissues　of　thigh　can　be　successfully　segmented　with　the　method.