A 1-D Convolutional Neural Network with Gradient Mapped Intensity Features for Detection of Mitosis in Histopathological Images
Akarsh Jagadeesha, Devaraj Verma C, Panita Wannapiroon, Junjiraporn Thongprasit · Applied Science and Engineering Progress · 2025
This paper proposes a mitosis detection algorithm that utilizes gradient-mapped intensity (GMI) features integrated into a one-dimensional convolutional neural network (1-D CNN) for the classification of mitotic cells in histopathological images. The proposed framework begins by preprocessing the input images through intensity compensation, followed by contrast enhancement using adaptive histogram equalization. Mitosis candidates are subsequently identified using adaptive thresholding techniques and morphological operations. From each detected candidate, GMI features are extracted through gradient estimation in both the x and y directions, construction of gradient histograms, and mapping of gradient magnitudes with corresponding intensity values. These features, derived from the red, green, and blue (RGB) channels, are used to train a 1-D CNN classifier that categorizes the inputs into two classes: mitosis and non-mitosis. The effectiveness of the proposed approach is evaluated using two benchmark datasets, ICPR 2012 and ICPR 2014, with performance measured via precision, recall, and F1-score metrics. The proposed model achieves an F1-score of 0.846, a recall of 0.859, and a precision of 0.863 on the ICPR 2012 dataset, demonstrating competitive performance compared to existing methods.