We compared non-negative matrix factorization (NMF) and convolution kernel compensation techniques for high-density electromyogram decomposition. The experimental data were recorded from nine healthy persons during controlled single degree of freedom (DOF) wrist flexion-extension, supination-pronation, and ulnar-radial deviation movements. We assembled the identified motor units and NMF components into three groups. Those active mostly during the first and the second movement direction per DOF were placed in the G1 and G3 groups, respectively. The remaining components were nonspecific for movement direction and were placed in the G2 group. In ulnar and radial deviation, the relative energies of identified cumulative motor unit spike trains (CSTs) and NMF components were similarly distributed among the groups. In other two movement types, the energy of NMF components in the G2 group was significantly larger than the energy of CSTs. We further performed a coherence analysis between CSTs and sums of NMF components in each group. Both decompositions demonstrated a solid match, but only at frequencies <3 Hz. At higher frequencies, the coherence hardly exceeded the value of 0.5. Potential reasons for these discrepancies include the negative impact of motor unit action potential shapes and noise on NMF decomposition.

We evaluated different muscle excitation estimation techniques, and their sensitivity to Motor Unit (MU) distribution in muscle tissue. For this purpose, the Convolution Kernel Compensation (CKC) method was used to identify the MU spike trains from High-Density Elec-troMyoGrams (HDEMG). Afterwards, Cumulative MU Spike Train (CST) was calculated by summing up the identified MU spike trains. Muscle excitation estimation from CST was compared to the recently introduced Cumulative Motor Unit Activity Index (CAI) and classically used Root-Mean-Square (RMS) amplitude envelop of EMG. To emphasize their dependence on the MU distribution further, all three muscle excitation estimates were used to calculate the agonist-antagonist co-activation index. We showed on synthetic HDEMG that RMS envelopes are the most sensitive to MU distribution (10 % dispersion around the real value), followed by the CST (7 % dispersion) and CAI (5 % dispersion). In experimental HDEMG from wrist extensors and flexors of post-stroke subjects, RMS envelopes yielded significantly smaller excitations of antagonistic muscles than CST and CAI. As a result, RMS-based co-activation estimates differed significantly from the ones produced by CST and CAI, illuminating the problem of large diversity of muscle excitation estimates when multiple muscles are studied in pathological conditions. Similar results were also observed in experimental HDEMG of six intact young males.

We analyzed muscle excitation estimation systematically by Non-negative matrix factorization (NMF) from surface electromyograms (EMG) during dynamic contractions of biceps brachii (BB) muscles. We used motor unit action potentials (MUAPs) estimated experimentally from surface EMGs during slow dynamic contractions of BB muscles in healthy young males, and convolved them by simulated motor unit firing patterns. Different uncorrelated muscle excitation and muscle shortening profiles were combined when generating the EMG signals from left and right BBs (64 channels per muscle). EMG signals were rectified, low-passed filtered and decomposed by NMF into 2, 3, 4 or 6 components. The identified NMF components demonstrated good separation of left and right BB activity, but relatively large sensitivity of NMF components to muscle shortening, especially at high levels of muscle excitation. When averaged across different numbers of identified NMF components at excitation levels ranging from 40 to 80 %, the average correlation coefficient between the NMF components and muscle shortening profiles was 0.45 ± 0.15. At contraction levels between 0 and 40 % these correlations decreased to 0.15 ± 0.09. Therefore, NMF components reflect both muscle excitation and muscle shortening profiles.

No abstract

Automated detection of ovarian follicles in ultrasound images is much appreciated when its effectiveness is comparable with the experts’ annotations. Today’s best methods estimate follicles notably worse than the experts. This paper describes the development of two-stage deeply-supervised 3D Convolutional Neural Networks (CNN) based on the established U-Net. Either the entire U-Net or specific parts of the U-Net decoder were replicated in order to integrate the prior knowledge into the detection. Methods were trained end-to-end by follicle detection, while transfer learning was employed for ovary detection. The USOVA3D database of annotated ultrasound volumes, with its verification protocol, was used to verify the effectiveness. In follicle detection, the proposed methods estimate follicles up to 2.9% more accurately than the compared methods. With our two-stage CNNs trained by transfer learning, the effectiveness of ovary detection surpasses the up-to-date automated detection methods by about 7.6%. The obtained results demonstrated that our methods estimate follicles only slightly worse than the experts, while the ovaries are detected almost as accurately as by the experts. Statistical analysis of 50 repetitions of CNN model training proved that the training is stable, and that the effectiveness improvements are not only due to random initialisation. Our deeply-supervised 3D CNNs can be adapted easily to other problem domains.