A Lightweight Algorithm for Classifying Ex Vivo Tissues Samples

TL;DR

A novel algorithm for classifying ex vivo tissue that comprises multi-channel bioimpedance analysis and a hardware neural network that can be integrated into the tip of a surgical margin assessment probe, for in vivo use during radical prostatectomy.

Abstract

In this paper, we present a novel algorithm for classifying ex vivo tissue that comprises multi-channel bioimpedance analysis and a hardware neural network. When implemented in a mixed-signal 180 nm CMOS process, the classifier has an estimated power budget of 39 mW and an area of 30 mm2. This means that the classifier can be integrated into the tip of a surgical margin assessment probe, for in vivo use during radical prostatectomy. We tested our classifier on digital phantoms of prostate tissue and also on an animal model of ex vivo bovine tissue. The classifier achieved an accuracy of 90% on the prostate tissue phantoms, and an accuracy of 84% on the animal model.

Figures (7)

• Figure 1: Proposed approach for prostate cancer detection involves 25-channel bioZ analysis, followed by a preprocessing stage and then a neural network. The neural network has the following architecture: Layer 1: 25-unit input stage; Layer 2: 16 units of a type of long short-term memory (AFUA); Layer 3: 2-unit fully-connected layer with sigmoid activation (FC1); Layer 4: 2-unit fully-connected layer with ReLU activation (FC2); Layer 5: 2-unit softmax output layer.
• Figure 2: Left panel: Example bioZ measurement. One set of bioZ data is collected by injecting a 1 mA$_{\rm pp}$ sinusoidal current ($I_{\rm inj}$) via a pair of electrodes (E33, E30) and measuring the induced voltages (V1, V2, etc.) with instrumentation amplifiers. These voltages are then processed to extract their amplitude and phase components. Voltage measurements are repeated for all possible current sink/source electrodes pair combinations (or 'current patterns') to form a 'frame' of bioZ data. Right panel: Electrode array configuration for our proposed surgical margin assessment probe. The voltage measurement electrodes are the small inner electrodes (i.e. electrodes 1 to 25). The current injection/sink electrodes are the larger electrodes on the probe's outer ring (i.e. electrodes 26 to 33). There are 8-choose-2=28 possible current patterns.
• Figure 3: Images of ex vivo bovine tissue samples with (left panel) mostly muscle tissue and (right panel) mixture of muscle and adipose tissue. The electrode array overlay (shown in red) indicates where a bioimpedance measurement was taken.
• Figure 4: AFUA schematic, comprising vector matrix multiplier ('VMM') and state update implemented as a gated current mirror. The current mirror incorporates activation functions that consume no extra power. The VMM is implemented using standard current mirror-based circuitry binas2016precise.
• Figure 5: Prostate digital phantom task. Accuracy curves for the full-precision (64-bit floating point) neural network model when evaluated on the training and validation sets of the prostate digital phantom data.