Automated bioelectric signal analysis has an important application in the wisdom medical care. In this work, we focus on ECG-signal and address a novel approach for cardiac arrhythmia diseases classification. We designed a novel analysis framework which extract different feature transformations from ECG signals. And we trained the SVM model for multi-feature to obtain the prediction. Finally, we tested our approach on the public database of MIT-BIH arrhythmia. And the results of experiments on the database demonstrate our model has better classification performance than other approaches.

Along with the great performance in diagnosing cardiovascular diseases, current stethoscopes perform unsatisfactorily in controlling undesired noise caused by the surrounding environment and detector operation. In this case, a low-noise-level heart sound system was designed to inhibit noise by a novel thorax-integration head with a flexible electric film. A hardware filter bank and wavelet-based algorithm were employed to enhance the recorded heart sounds from the system. In the experiments, we used the new system and the 3M™ Littmann® Model 3200 Electronic Stethoscope separately to record heart sounds in different noisy environments. The results illustrated that the average estimated noise ratio represented 21.26% and the lowest represented only 12.47% compared to the 3M stethoscope, demonstrating the better performance in denoising ability of this system than state-of-the-art equipment. Furthermore, based on the heart sounds recorded with this system, some diagnosis results were achieved from an expert and compared to echocardiography reports. The diagnoses were correct except for two uncertain items, which greatly confirmed the fact that this system could reserve complete pathological information in the end.

Noisy labels are often encountered in datasets, but learning with them is challenging. Although natural discrepancies between clean and mislabeled samples in a noisy category exist, most techniques in this field still gather them indiscriminately, which leads to their performances being partially robust. In this paper, we reveal both empirically and theoretically that the learning robustness can be improved by assuming deep features with the same labels follow a student distribution, resulting in a more intuitive method called student loss. By embedding the student distribution and exploiting the sharpness of its curve, our method is naturally data-selective and can offer extra strength to resist mislabeled samples. This ability makes clean samples aggregate tightly in the center, while mislabeled samples scatter, even if they share the same label. Additionally, we employ the metric learning strategy and develop a large-margin student (LT) loss for better capability. It should be noted that our approach is the first work that adopts the prior probability assumption in feature representation to decrease the contributions of mislabeled samples. This strategy can enhance various losses to join the student loss family, even if they have been robust losses. Experiments demonstrate that our approach is more effective in inaccurate supervision. Enhanced LT losses significantly outperform various state-of-the-art methods in most cases. Even huge improvements of over 50% can be obtained under some conditions.

Gastrointestinal stromal tumor (GIST) is the most common malignant tumor originating from interstitial cells in the gastrointestinal tract. Different grades require various surgical interventions and adjuvant treatments, which are closely linked to the patient’s prognosis. The current clinical risk stratification method relies heavily on the identification and counting of mitotic figures, which serve as important criteria. However, manual evaluation of pathological slides in clinical practice is often limited by the shortage of experienced clinicians and the subjectivity in interpreting results. Therefore, in this paper, we propose an enhanced YOLOv8 network framework for the automatic detection of mitotic cells in GIST. We substituted the backbone network with VanillaNet, known for its simplified model complexity in feature extraction. This change facilitated the identification of specific targets and improved network performance. Additionally, we introduced the Advanced Feature Pyramid Network (AFPN) to further enhance the model’s accuracy. Experimental results show that the proposed model achieved an accuracy of 0.816, a recall rate of 0.858, and an F1-score of 0.837 on the test dataset. It demonstrates superior efficacy in identifying mitotic cells, outperforming the original YOLOv8 model in overall performance. This augmented model has the potential to significantly reduce reading time while ensuring consistent diagnostic results, thereby greatly improving diagnostic efficiency. Future large-scale validation is necessary for the clinical adoption of this model.