Diabetic foot ulcers (DFUs) result in amputation of lower limbs or feet without timely assessment and treatment. The assessment of DFUs is performed by the diagnosis of DFU ischaemia and infection. In this work, a new deep convolutional neural network (CNN) based approach (ResKNet) is proposed to perform such assessment. The proposed network consists of a series of unique residual blocks of 2D convolution, batch normalization, and LeakyReLU with skip connections. It has been experimentally observed that a shallower network with 4 such unique residual blocks (Res4Net) can achieve promising results in ischaemia recognition, with an impressive AUC value of 0.9968. However in case of infection recognition, a network with 7 such residual blocks (Res7Net) achieved the best performance, with an AUC value of 0.8890. Such improvements to the current state-of-the-art results suggests that the proposed approach can provide significant help to medical experts, for automatic assessment of DFU.

中文翻译：

---

糖尿病足溃疡缺血和感染的识别：基于深度卷积神经网络的方法

糖尿病足溃疡 (DFU) 导致下肢或足部截肢，而没有及时评估和治疗。DFU 的评估是通过 DFU 缺血和感染的诊断来进行的。在这项工作中，提出了一种新的基于深度卷积神经网络 (CNN) 的方法 (ResKNet) 来执行此类评估。所提出的网络由一系列独特的 2D 卷积残差块、批量归一化和带有跳跃连接的 LeakyReLU 组成。实验观察到，具有 4 个此类独特残差块的较浅网络（Res4Net）可以在缺血识别方面取得可喜的结果，其 AUC 值为 0.9968，令人印象深刻。然而，在感染识别的情况下，具有 7 个此类残差块的网络（Res7Net）取得了最佳性能，AUC 值为 0.8890。