Defocusing is used in bright-field image processing in order to increase image contrast. Moreover, defocused images can be used to solve the transport of intensity equation (TIE) and obtain physical light phase. Recently, it was shown that the monogenic local features of an axial intensity derivative passed through a specific low-pass filter can be used to improve cell segmentation. In this paper, we show that the TIE solution and the low-pass monogenic local phase (LMLP) can be successfully employed for improving joint learning of adherent and suspended cell detection. A state-of-the-art approach for cell detection on defocused images reported 10.4% decrease in F-measure of suspended cell detection when trained on both adherent and suspended cell lines compared to the case when training was done for each cell line separately. Using TIE solution for feature extraction instead of a defocused image, joint training was drastically improved and the aforementioned difference in F-measure was reduced to 2%. LMLP, achieved approximately the same result, though a bit inferior.