The Human Epidermal Growth Factor Receptor 2 (HER2) is one of the aggressive subtypes of breast cancer. The HER2 status decides the requirement of breast cancer patients to receive HER2-targeted therapy. The HER2 testing involves combining Immunohistochemistry (IHC) screening, followed by fluorescence in situ hybridization (FISH) for cases where IHC results are equivocal. These tests may involve multiple trials, are time intensive, and tend to be more expensive for certain classes of people. Hematoxylin and Eosin (HE) staining is employed for visualizing general tissue morphology and is a routine, cost-effective method. In this study, we introduce a novel automated class-wise weighted average ensemble deep learning algorithm at the decision level. The proposed algorithm fuses three pre-trained deep-learning models at the decision level by assigning a weight to each class based on their performance of the model to classify the HE-stained breast histopathology images into multi-class HER2 statuses as HER2-0+, HER2-1+, HER2-2+, and HER2-3+. The class-wise weight allocation to the base classifiers is one of the key features of the proposed algorithm. The presented framework surpasses all the existing methods currently employed on the Breast Cancer Immunohistochemistry (BCI) dataset, establishing itself as a dependable approach for detecting HER2 status from HE-stained images. This study highlights the robustness of the proposed algorithm as well as the sufficient information encapsulated within HE-stained images for the effective detection of the HER2 protein present in breast cancer. Therefore, the proposed method possesses the potential to sideline the need for IHC laboratory tests, which hoard time and money.

Vertebral fracture (VF) may induce spinal cord injury that can lead to serious consequences which eventually may paralyze the entire or some parts of the body depending on the location and severity of the injury. Diagnosis of VFs is crucial at the initial stage, which may be challenging because of the subtle features, noise, and homogeneity present in the computed tomography (CT) images. In this study, Wide ResNet-40, DenseNet-121, and EfficientNet-B7 are chosen, fine-tuned, and used as base models, and a Bayesian-based probabilistic ensemble learning method is proposed for fracture detection in cervical spine CT images. The proposed method considers the prediction’s uncertainty of the base models and combines the predictions obtained from them, to improve the overall performance significantly. This method assigns weights to the base learners, based on their performance and confidence about the prediction. To increase the robustness of the proposed model, custom data augmentation techniques are performed in the preprocessing step. This work utilizes 15,123 CT images from the RSNA-2022 C-spine fracture detection challenge and demonstrates superior performance compared to the individual base learners, and the other existing conventional ensemble methods. The proposed model also outperforms the best state-of-the-art (SOTA) model by 1.62%, 0.51%, and 1.29% in terms of accuracy, specificity, and sensitivity, respectively; furthermore, the AUC score of the best SOTA model is lagging by 5%. The overall accuracy, specificity, sensitivity, and F1-score of the proposed model are 94.62%, 93.51%, 95.29%, and 93.16%, respectively.

Accurate segmentation of vertebrae in computed tomography (CT) images possess serious challenges due to the irregular vertebral boundaries, low contrast and brightness, and noise in CT scans. This study presents a novel channel attention-based EfficientNetB7-UNet (CAEB7-UNet) method to address this complex task effectively. The proposed model introduces an upgraded ReLU-based channel attention module (CAM) in the skip connection which restrains the nonessential attributes by suppressing them and accentuates the relevant features by emphasizing them to boost the overall segmentation performance. In this work, an improved EfficientNetB7 is employed as the encoder for feature extraction, the fusion of local and global features is enhanced through the upgraded CAM in skip connection, and the up-sampling is performed in the decoder. Further, the model is optimized by incorporating hyperparameter optimization, specifically, hybrid learning rate scheduler strategies, along with the AdamW optimizer and custom data augmentation. A total of 34,782 CT images obtained from the RSNA-2022 cervical spine fracture detection challenge is utilized in this study. The proposed model achieves outstanding performance, yielding a dice score index (DSI) of 96.14% and mean intersection over union (mIoU) of 91.46%. Moreover, a comparative performance analysis of CAEB7-UNet with two state-of-the-art models is carried out on the same dataset. Our approach outperforms both the models, with the best one by 8.1%, 6.73%, 12.7%, and 11.98% in terms of DSI, mIoU, precision, and F1-score respectively. Additionally, it requires merely 0.38 seconds to generate the segmentation mask of a single slice of a CT scan.

Human epidermal growth factor receptor 2 (HER2) is a critical gene that serves as a receptor to transmit signals for aggressive cell division in cancer cells. Hence, testing of HER2 is important in treatment to indicate candidates for HER2-targeted therapy. However, in the current gold standard, i.e., the immunohistochemistry (IHC) test, the scoring is based on the pathologist’s analysis, which has an inter- and intra-observer variation chance due to variability in staining assessment. Automating HER2 scoring using hematoxyline eosin (HE) stained images can overcome these limitations, providing more accurate and consistent results, thereby reducing healthcare costs and enhancing patient outcomes. In this work, we have presented an automated framework for classifying HER2 scores of breast cancer using HE-stained images. The developed framework uses three fine-tuned deep learning models, namely GoogLeNet, ResNet-50, and Vision Transformer (ViT). It applies the proposed hybrid weighted individual voting ensemble (WIVE) to combine the confidence scores of all the constituent models. This approach comprises two independent techniques: the Model-Specific Weights Optimization (MSWO), customizing weights for individual models, and the Class-Specific Weights Optimization (CSWO), fine-tunes weights for specific classes using Probabilistic model-building genetic algorithms (PMBGAs). The proposed framework surpasses the existing methods in the literature. The CSWO approach achieves an accuracy of 99.42% and a precision of 99.54%, while the MSWO approach attains an accuracy of 99.07% and a precision of 99.21%. This study outlines an economically feasible and efficient prognostic model with the potential to provide clinically significant inputs. The use of this algorithm might offer a possibility for the replacement of IHC testing, minimizing the variability in HER2 scoring, as well as simplifying the diagnostic process.

Accurate identification of vertebral fracture (VF) regions in computed tomography (CT) images is crucial for surgeons prior to treatment planning, but remains challenging due to irregular vertebral boundaries, low contrast, noise, and image unevenness. Recent advancements in deep learning have shown promising results compared to conventional manual diagnosis methods in detecting anomalies and segmenting regions of interest in medical imaging. In this study, a deep learning model, enhanced EfficientNetB7 U-Net (EEB7-UNet), is proposed to segment the fractured cervical vertebrae. It includes custom data augmentation to increase the data size and a hybrid learning rate scheduler strategy technique for faster convergence, which increases the generalizability and robustness of the model. The proposed model achieved an improved dice score index of 95.53% and Jaccard coefficient index of 93.85% on the test dataset. Furthermore, the EEB7-UNet has emerged as a moderate size with 98.6 MB. The approach yields superior performance in terms of dice score index and Jaccard coefficient index compared to the other state of the art convolutional neural network (CNN) used as an encoder in the U-Net. This research also compared the performance of the proposed model with three other studies in similar contexts, reported in the literature.