DeepCon⁺ based Deep Learning Architecture for Histopathology Image Classification

Histopathological image classification is challenging due to substantial intra-class variability and similarities between tissue categories. This paper presents a DeepCon⁺ framework for colorectal cancer tissue classification. First, a pretrained Xception network is used to extract deep feature vectors from the images. Principal Component Analysis (PCA) is used to reduce the feature dimensionality while retaining 95% of the original information. Subsequently, class-wise K-means clustering decomposes the eight original tissue classes into 16 subclasses to represent intra-class variations more effectively. A pretrained Xception-based CNN is then fine-tuned using transfer learning to classify the decomposed subclasses. This coarse-learning stage enables the model to learn more discriminative representations of heterogeneous tissue patterns. The overall framework aims to improve feature representation and classification by combining deep feature extraction, dimensionality reduction, class wise clustering, and subclass-level learning. Its effectiveness can be assessed using standard classification metrics, including accuracy, precision, recall, and F1-score. The experimental analysis performed on the colorectal histopathology dataset reveals that the proposed Deep Con⁺ framework markedly surpasses the initial Deep Con model and traditional CNN benchmarks, attaining enhanced classification accuracy and increased F1-score for all tissue categories.